Papers with supervised fine-tuning
Advancing Language Models through Instruction Tuning: Recent Progress and Challenges (2025.emnlp-tutorials)
Copied to clipboard
| Challenge: | tutorial addresses three critical questions within the field of instruction tuning: (1) What are the current focal points in instruction tuning research? (2) What are best practices in training an instruction-following model? (3) What new challenges have emerged? |
| Approach: | This tutorial presents a systematic overview of recent advances in instruction tuning. |
| Outcome: | The tutorial covers different stages in model training: supervised fine-tuning, preference optimization, and reinforcement learning. |
RAVEN: Robust Advertisement Video Violation Temporal Grounding via Reinforcement Reasoning (2025.acl-industry)
Copied to clipboard
| Challenge: | Existing methods for detecting ads video violations lack precise temporal grounding, noisy annotations, and limited generalization. |
| Approach: | They propose a framework that integrates curriculum reinforcement learning with large language models to enhance reasoning and cognitive capabilities for violation detection. |
| Outcome: | The proposed framework achieves superior performance in violation category accuracy and temporal interval localization. |
WorkForceAgent-R1: Incentivizing Reasoning Capability in LLM-based Web Agents via Reinforcement Learning (2026.findings-eacl)
Copied to clipboard
| Challenge: | Existing web agents relying on supervised fine-tuning struggle with generalization and robustness due to insufficient reasoning capabilities when handling the inherently dynamic nature of web interactions. |
| Approach: | They propose a large language model-empowered web agent that trains using a rule-based reinforcement learning framework to enhance single-step reasoning and planning for business-oriented web navigation tasks. |
| Outcome: | The proposed agent outperforms baseline LLM-based agents on the WorkArena benchmark by 10.26–16.59%. |
Improving Tool Retrieval by Leveraging Large Language Models for Query Generation (2025.coling-industry)
Copied to clipboard
| Challenge: | Large Language Models (LLMs) have shown great promise in common sense language understanding, conversational fluency, and reasoning. |
| Approach: | They propose to use Large Language Models to generate a retrieval query and embed it into the prompt to find relevant tools via a nearest-neighbor search. |
| Outcome: | The proposed method improves retrieval for in-domain (seen tools) and out-of-domain settings. |
The Pragmatic Mind of Machines: Tracing the Emergence of Pragmatic Competence in Large Language Models (2026.eacl-long)
Copied to clipboard
| Challenge: | Current large language models (LLMs) have demonstrated emerging capabilities in social intelligence tasks, including implicature resolution and theory-of-mind reasoning. |
| Approach: | They introduce a dataset grounded in the pragmatic concept of alternatives to evaluate whether large language models can accurately infer nuanced speaker intentions. |
| Outcome: | The proposed model can infer nuanced speaker intentions by inferring the speaker’s intended meaning and explaining when and why a speaker would choose one utterance over its alternative. |
Best Practices for Distilling Large Language Models into BERT for Web Search Ranking (2025.coling-industry)
Copied to clipboard
| Challenge: | Recent studies have highlighted the potential of Large Language Models (LLMs) as zero-shot relevance rankers. |
| Approach: | They propose to use a ranking loss to transfer ranking knowledge from LLMs to smaller models like BERT. |
| Outcome: | The proposed model has been successfully integrated into a commercial web search engine as of February 2024. |
AgentCPM-GUI: Building Mobile-Use Agents with Reinforcement Fine-Tuning (2025.emnlp-demos)
Copied to clipboard
Zhong Zhang, Yaxi Lu, Yikun Fu, Yupeng Huo, Shenzhi Yang, Yesai Wu, Han Si, Xin Cong, Haotian Chen, Yankai Lin, Xie Xie, Wei Zhou, Wang Xu, Zhou Su, Zhongwu Zhai, Xiaoming Liu, null Meiyudong, Jianming Xu, Hongyan Tian, Chongyi Wang, Chi Chen, Yuan Yao, Zhiyuan Liu, Maosong Sun
| Challenge: | Large language model agents have enabled GUI-based automation, but their deployment is limited by noisy data, poor generalization, and lack of support for non-English GUIs. |
| Approach: | They propose an 8B-parameter GUI agent built for robust and efficient on-device GUI interaction. |
| Outcome: | The proposed GUI agent achieves promising performance on five public benchmarks and proposed Chinese benchmark CAGUI. |
How Abilities in Large Language Models are Affected by Supervised Fine-tuning Data Composition (2024.acl-long)
Copied to clipboard
Guanting Dong, Hongyi Yuan, Keming Lu, Chengpeng Li, Mingfeng Xue, Dayiheng Liu, Wei Wang, Zheng Yuan, Chang Zhou, Jingren Zhou
| Challenge: | supervised fine-tuning (SFT) is a technique used to enhance multiple abilities in large language models. |
| Approach: | They propose to study the interplay of data composition between mathematical reasoning, code generation, and general human-aligning abilities during supervised fine-tuning. |
| Outcome: | The proposed model improves math reasoning and code generation with increasing data amount . the proposed model size and SFT strategies can be used to learn multiple skills with different scaling patterns. |
TL-Training: A Task-Feature-Based Framework for Training Large Language Models in Tool Use (2025.findings-emnlp)
Copied to clipboard
Junjie Ye, Yilong Wu, Sixian Li, Yuming Yang, Zhiheng Xi, Tao Gui, Qi Zhang, Xuanjing Huang, Peng Wang, Zhongchao Shi, Jianping Fan, Zhengyin Du
| Challenge: | a new approach to training large language models (LLMs) overlooks task-specific characteristics in tool use, leading to performance bottlenecks. |
| Approach: | They propose a task-feature-based framework that mitigates the effects of suboptimal training data . they use a dataset to train large-scale LLMs and a reward mechanism tailored to error categories . |
| Outcome: | The proposed framework matches or surpasses open- and closed-source LLMs in tool-use performance using only 1,217 training data points. |
Lacuna Reconstruction: Self-Supervised Pre-Training for Low-Resource Historical Document Transcription (2022.findings-naacl)
Copied to clipboard
| Challenge: | Document transcription models are limited by extremely varied style and content across domains. |
| Approach: | They propose a self-supervised approach for learning rich visual representations for both handwritten and printed historical document transcription using a heterogeneous set of handwritten Islamicate manuscript images and early modern English printed documents. |
| Outcome: | The proposed model improves on a supervised model with as few as 30 line image transcriptions on two languages with a single line of image training. |
Deploying a Retrieval based Response Model for Task Oriented Dialogues (2022.emnlp-industry)
Copied to clipboard
| Challenge: | a task-oriented dialogue system needs high conversational capability and can be easily adaptable to changing situations. |
| Approach: | They propose a retrieval-based conversational model that can rank a large set of responses . they propose supervised training and fine-tuning on limited data collected through a human-in-the-loop platform . |
| Outcome: | The proposed model can scale to rank a large set of responses in real-world situations. |
MGPO: Thinking with Images via Multi-Turn Grounding-Based Reinforcement Learning (2026.findings-acl)
Copied to clipboard
| Challenge: | State-of-the-art large multimodal models face challenges when processing high-resolution images, as these inputs are converted into enormous visual tokens, many of which are irrelevant to the downstream task. |
| Approach: | They propose a multi-turn grounding-based policy optimization framework that enables LMMs to iteratively focus on key visual regions by automatically cropping sub-images based on model-predicted grounding coordinates within a multiple-turn conversation framework. |
| Outcome: | The proposed framework improves on Qwen2.5-VL-7B with 21K samples and surpasses OpenAI’s o1 and GPT-4o models on the out-of-distribution (OOD) V* Bench. |
Distilling Text Style Transfer With Self-Explanation From LLMs (2024.naacl-srw)
Copied to clipboard
| Challenge: | Text Style Transfer (TST) aims to alter the style of text while preserving its core content. |
| Approach: | They propose a framework that leverages large language models alongside chain-of-thought prompting to facilitate TST. |
| Outcome: | The proposed framework surpasses supervised fine-tuning and knowledge distillation methods in low-resource settings. |
Call, Reward, Repeat: Advancing Dialog State Tracking with GRPO and Function Calling (2026.eacl-srw)
Copied to clipboard
| Challenge: | Recent advances in Large Language Models (LLMs) have notably enhanced task-oriented dialogue systems, particularly in Dialogue State Tracking (DST). |
| Approach: | They propose a group-relative policy optimization method that guides LLMs toward improved DST accuracy even under low-resource conditions. |
| Outcome: | The proposed method improves on established DST benchmarks while using significantly reduced out-of-domain training data. |
Better Literary Translation: A Multi-Aspect Data Generation and LLM Training Approach (2026.acl-industry)
Copied to clipboard
| Challenge: | Literary translation requires balancing expression fluency with literary effect due to the scarcity of high-quality training data and the difficulty of capturing nuanced quality trade-offs. |
| Approach: | They propose a multi-aspect iterative refinement framework that generates high-quality translation references and preference data through specialized LLM translators. |
| Outcome: | The proposed models outperform the ground truth for SFT by 8.65 CEA100 points while leveraging an explicit reward model for GRPO yields an additional 1.51 point improvement. |
LEAF: Learning and Evaluation Augmented by Fact-Checking to Improve Factualness in Large Language Models (2025.emnlp-industry)
Copied to clipboard
| Challenge: | Large language models (LLMs) struggle with factual accuracy in knowledge-intensive domains like healthcare. |
| Approach: | They propose a framework for improving LLM factuality in medical question answering . RAFE, Fact-Check-then-RAG and Learning from Fact Check are components . |
| Outcome: | Experimental results show that LEAF outperforms Factcheck-GPT in detecting inaccuracies and corrects errors without labeling . the framework provides a scalable solution for industrial applications requiring high factuality scores. |
Towards Hierarchical Multi-Step Reward Models for Enhanced Reasoning in Large Language Models (2026.findings-acl)
Copied to clipboard
Teng Wang, Jiang Zhangyi, Zhenqi He, Hailei Gong, Shenyang Tong, Wenhan Yang, Zeyu Li, Yanan Zheng, Zifan He, Zewen Ye, Shengjie Ma, Jianping Zhang
| Challenge: | Existing Process Reward Models (PRMs) are vulnerable to reward hacking and require expensive, large-scale annotation of reasoning steps. |
| Approach: | They propose a reward model approach which evaluates both individual and consecutive reasoning steps from fine-grained and coarse-grounded level. |
| Outcome: | Empirical results show that the proposed model performs better than existing PRMs and is more robust than existing models. |
Model Merging for Knowledge Editing (2025.acl-industry)
Copied to clipboard
Zichuan Fu, Xian Wu, Guojing Li, Yingying Zhang, Yefeng Zheng, Tianshi Ming, Yejing Wang, Wanyu Wang, Xiangyu Zhao
| Challenge: | Existing knowledge editing approaches struggle with sequential editing scenarios and harm the general capabilities of the model. |
| Approach: | They propose a framework that combines robust supervised fine-tuning and model merging for knowledge editing to combine supervised and supervised learning. |
| Outcome: | The proposed approach outperforms existing methods in sequential editing while preserving the original performance of the model. |
Rescue: Ranking LLM Responses with Partial Ordering to Improve Response Generation (2024.acl-srw)
Copied to clipboard
| Challenge: | Customizing LLMs for a specific task involves separating high-quality responses from lower-quality ones. Obtaining a large volume of expert-annotated data is costly for most tasks. |
| Approach: | They propose a method that trains the model to prioritize the best responses from a pool of candidates created for a task using ranking metrics. |
| Outcome: | The proposed method is more robust, less sensitive to noise, and can be achieved with limited human annotations or through heuristic methods. |
BigTokDetect: A Clinically-Informed Vision–Language Modeling Framework for Detecting Pro-Bigorexia Videos on TikTok (2026.eacl-long)
Copied to clipboard
Minh Duc Chu, Kshitij Pawar, Zihao He, Roxanna Sharifi, Ross M. Sonnenblick, Magdalayna Curry, Laura DAdamo, Lindsay Young, Stuart Murray, Kristina Lerman
| Challenge: | Social media platforms face escalating challenges in detecting harmful content that promotes muscle dysmorphic behaviors and cognitions (bigorexia). |
| Approach: | They propose a framework for detecting pro-bigorexia content on TikTok using an expert-annotated multimodal benchmark dataset of over 2,200 Tiktok videos labeled by clinical psychiatrists. |
| Outcome: | The proposed framework improves on fine-grained subcategories while commercial models achieve the highest accuracy on primary categories. |
Complexity-aware fine-tuning (2026.findings-eacl)
Copied to clipboard
| Challenge: | General-purpose Large Language Models (LLMs) are often fine-tuned through supervised fine- tuning (SFT) to enhance performance in specific domains. |
| Approach: | They propose a novel approach that uses reasoning only for complex data identified by entropy to refine large language models. |
| Outcome: | The proposed model outperforms the standard SFT approach while using 81% less data. |
SpatialMath: Spatial Comprehension-Infused Symbolic Reasoning for Mathematical Problem-Solving (2026.findings-eacl)
Copied to clipboard
| Challenge: | Current models struggle to accurately decompose intricate visual inputs and connect perception with structured reasoning, leading to suboptimal performance. |
| Approach: | They propose a Spatial Comprehension-Infused Symbolic Reasoning Framework to integrate spatial representations into structured symbolic reasoning chains. |
| Outcome: | The proposed framework outperforms existing models in vision-intensive mathematical problems. |
DecIF: Improving Instruction-Following through Decomposition (2026.acl-long)
Copied to clipboard
| Challenge: | Existing approaches to obtain high-quality instruction-following data rely heavily on existing documents and existing methods. |
| Approach: | They propose a data synthesis framework, DecIF, which automatically generates accurate and diverse instruction-following data from scratch for supervised fine-tuning and reinforcement learning. |
| Outcome: | Extensive experiments show that the proposed framework can synthesize accurate instruction-following data for both SFT and RL paradigms compared to baselines. |
PAFT: Prompt-Agnostic Fine-Tuning (2025.emnlp-main)
Copied to clipboard
| Challenge: | Prompt-agnostic fine-tuning (PAFT) improves performance by reducing overfitting to specific prompts. |
| Approach: | They propose a method that enhances robustness through dynamic prompt variation during training. |
| Outcome: | The proposed method achieves higher generalization accuracy on unseen prompts than standard methods with similar training efficiency. |
Prompt2Model: Generating Deployable Models from Natural Language Instructions (2023.emnlp-demo)
Copied to clipboard
| Challenge: | Large language models (LLMs) are a step backward from traditional special-purpose NLP models . they require extensive computational resources for deployment and can be gated behind APIs . |
| Approach: | They propose a general-purpose method that takes a natural language task description and uses it to train a special-purpose model. |
| Outcome: | The proposed method outperforms a strong LLM by 20% while being 700 times smaller. |
SparkRA: A Retrieval-Augmented Knowledge Service System Based on Spark Large Language Model (2024.emnlp-demo)
Copied to clipboard
Dayong Wu, Jiaqi Li, Baoxin Wang, Honghong Zhao, Siyuan Xue, Yanjie Yang, Zhijun Chang, Rui Zhang, Li Qian, Bo Wang, Shijin Wang, Zhixiong Zhang, Guoping Hu
| Challenge: | Large language models (LLMs) have shown remarkable achievements across various language tasks. |
| Approach: | They propose a scientific literature LLM and a knowledge service system based on it . they collect scientific literature and then pre-train it using autoregressive training . |
| Outcome: | The proposed system provides literature investigation, paper reading, and academic writing functions. |
Reward Engineering for Generating Semi-structured Explanation (2024.findings-eacl)
Copied to clipboard
| Challenge: | Unstructured natural language explanations lack a comprehensive explanation mechanism to verify a model's true reasoning capabilities. |
| Approach: | They propose a reward engineering method which uses semi-structured explanations to verify a model's true reasoning capabilities. |
| Outcome: | The proposed method achieves new state-of-the-art on two semi-structured explanation generation benchmarks (ExplaGraph and COPA-SSE) . |
Learning to Translate Ambiguous Terminology by Preference Optimization on Post-Edits (2025.emnlp-industry)
Copied to clipboard
| Challenge: | Ambiguous terminology can make translation difficult, especially in corporate contexts. |
| Approach: | They propose to learn how to disambiguate terminology based on human post-edits . they use preference optimization to optimize for correctness using the term post-Edit . |
| Outcome: | The proposed framework improves term accuracy over a translation oriented LLM without significant losses in COMET score. |
BIDER: Bridging Knowledge Inconsistency for Efficient Retrieval-Augmented LLMs via Key Supporting Evidence (2024.findings-acl)
Copied to clipboard
| Challenge: | Large language models (LLMs) have demonstrated efficacy in knowledge-intensive tasks such as open-domain QA, but inconsistencies between retrieval knowledge and the necessary knowledge for LLMs, leading to a decline in LLM’s answer quality. |
| Approach: | They propose a retrieval-augmented large language model that refines retrieval documents into Key Supporting Evidence (KSE) through knowledge synthesis, supervised fine-tuning, and preference alignment. |
| Outcome: | The proposed model outperforms existing methods in five datasets and boosts answer quality by 7% while reducing input content length by 80%. |
A Novel Paradigm Boosting Translation Capabilities of Large Language Models (2024.findings-naacl)
Copied to clipboard
| Challenge: | Existing studies on LLMs focused on supervised fine-tuning but their effectiveness has been limited. |
| Approach: | They propose a paradigm consisting of three stages: Secondary Pre-training using extensive monolingual data, Continual Pre- training with interlinear text format documents, and Leveraging source-language consistent instruction for supervised fine-tuning. |
| Outcome: | The proposed approach surpasses previous work and achieves superior performance compared to models such as NLLB-54B(CITATION) and GPT3.5-text-davinci-003. |
Aligning What LLMs Do and Say: Towards Self-Consistent Explanations (2026.findings-acl)
Copied to clipboard
| Challenge: | Large language models (LLMs) are often prompted to produce natural language explanations, but the features driving the answer are often different from those emphasized in their explanations. |
| Approach: | They propose a large-scale benchmark linking model decisions with diverse explanations and attribution vectors across datasets, methods, and model families to address this gap. |
| Outcome: | The proposed model generates an answer where the word NLP in the prompt has high feature importance. |
Skill-Aware Data Selection and Fine-Tuning for Data-Efficient Reasoning Distillation (2026.acl-short)
Copied to clipboard
| Challenge: | Large reasoning models such as DeepSeek-R1 and their distilled variants achieve impressive performance on complex reasoning tasks, yet their costs remain substantial. |
| Approach: | They propose a skill-centric distillation framework that efficiently transfers reasoning ability to weaker models with two components: (1) Skill-based data selection, which prioritizes examples targeting the student model’s weaker skills, and (2) Skillaware fine-tuning, which encourages explicit skill decomposition during problem solving. |
| Outcome: | The proposed framework surpasses baselines on Qwen3-4B and Qwend3-8B and focuses on skills emphasized during training. |
Exploring Self-supervised Logic-enhanced Training for Large Language Models (2024.naacl-long)
Copied to clipboard
| Challenge: | Traditional attempts to enhance the logical reasoning abilities of language models often rely on supervised fine-tuning, limiting their generalization to new tasks or domains. |
| Approach: | They propose a framework for integrating logical reasoning capabilities into LLMs and activating them via in-context learning. |
| Outcome: | The proposed framework achieves comparable results to existing models on three language understanding benchmarks. |
Improve Language Model and Brain Alignment via Associative Memory (2025.findings-acl)
Copied to clipboard
| Challenge: | Existing studies have shown that associative memory is essential for language comprehension and comprehension. |
| Approach: | They propose to integrate associative memory into language models to improve alignment . they find alignment is improved in brain regions closely related to associativ memory processing . |
| Outcome: | The proposed model improves in brain regions closely related to associative memory processing. |
Reflect-RL: Two-Player Online RL Fine-Tuning for LMs (2024.acl-long)
Copied to clipboard
| Challenge: | supervised fine-tuning (SFT) on a limited offline dataset does not yield good performance. |
| Approach: | They propose a two-player system to fine-tune an LM using SFT and online RL . they use negative example generation to enhance error-correction ability of the reflection model . |
| Outcome: | The proposed system outperforms SFT and online RL without reflection on a GPT-2 XL 1.56B model. |
Towards Reliable Large Audio Language Model (2025.findings-acl)
Copied to clipboard
Ziyang Ma, Xiquan Li, Yakun Song, Wenxi Chen, Chenpeng Du, Jian Wu, Yuanzhe Chen, Zhuo Chen, Yuping Wang, Yuxuan Wang, Xie Chen
| Challenge: | Recent advances in large audio language models (LALMs) have demonstrated impressive results and promising prospects in universal understanding and reasoning across speech, music, and general sound. |
| Approach: | They propose to use training-free and training-based methods to enhance LALM reliability to different extents. |
| Outcome: | The proposed methods improve the reliability of large audio language models to different extents. |
GTA: Supervised-Guided Reinforcement Learning for Text Classification with Large Language Models (2025.findings-emnlp)
Copied to clipboard
| Challenge: | Reinforcement learning fine-tuning methods suffer from inefficient exploration and slow convergence . supervised fine- tuning methods have limited performance ceiling and less solid theoretical foundation . |
| Approach: | They propose a Guess-Think-Answer framework that combines supervised and supervised learning in a unified training paradigm. |
| Outcome: | The proposed framework outperforms both standalone SFT and RL training models on three text classification benchmarks. |
RespiraMFM: A Multimodal Foundation Model with Contrastive Audio-Language Alignment for Respiratory Disease Identification (2026.acl-long)
Copied to clipboard
| Challenge: | Existing models for respiratory diseases rely on audio inputs, but they lack generalizability and diagnostic precision. |
| Approach: | They propose a multimodal foundation model that integrates respiratory sounds with medical history and symptoms to enhance diagnostic accuracy and disease detection capabilities. |
| Outcome: | The proposed model improves AUROC and zero-shot tasks across five respiratory diseases using real-world datasets. |
VaseVQA: Multimodal Agent and Benchmark for Ancient Greek Pottery (2026.findings-eacl)
Copied to clipboard
Jinchao Ge, Tengfei Cheng, Biao Wu, Zeyu Zhang, Shiya Huang, Judith Bishop, Gillian Shepherd, Meng Fang, Ling Chen, Yang Zhao
| Challenge: | MLLMs that use domain-specific data are limited in understanding cultural heritage artifacts such as ancient Greek pottery . supervised fine-tuning improves adaptation to domain knowledge, but it struggles with deeper reasoning tasks. |
| Approach: | They propose a visual question-answer tool that augments SFT with reinforcement learning using verifiable rewards. |
| Outcome: | The proposed model outperforms baseline models on reasoning-intensive questions on ancient Greek pottery. |
EasyDistill: A Comprehensive Toolkit for Effective Knowledge Distillation of Large Language Models (2025.emnlp-demos)
Copied to clipboard
| Challenge: | In this paper, we present EasyDistill, a comprehensive toolkit designed for effective black-box and white-box knowledge distillation (KD) of large language models. |
| Approach: | They propose a toolkit for effective black-box and white-box knowledge distillation (KD) of large language models (LLMs). |
| Outcome: | The framework offers data synthesis, supervised fine-tuning, ranking optimization, and reinforcement learning techniques specifically tailored for KD scenarios. |
Language Models Are Poor Learners of Directional Inference (2022.findings-emnlp)
Copied to clipboard
| Challenge: | Existing datasets fail to test directionality and are infested by artefacts that can be learnt as proxy for entailments, yielding over-optimistic results. |
| Approach: | They propose a benchmark for directional predicate entailments that is extrinsic to existing training sets. |
| Outcome: | The proposed model is incompetent on directional predicate entailments, compared to engorgement graphs, but limited by sparsity. |
Little Giants: Synthesizing High-Quality Embedding Data at Scale (2025.naacl-long)
Copied to clipboard
| Challenge: | Synthetic data generation is an increasingly popular way of training models without the need for large, manually labeled datasets. |
| Approach: | They propose a framework that aligns open-source small models to efficiently generate large-scale embedding data. |
| Outcome: | The proposed framework outperforms state-of-the-art embedding models by using only 1/10 of the GPT API calls. |
Improving Factuality in Clinical Abstractive Multi-Document Summarization by Guided Continued Pre-training (2024.naacl-short)
Copied to clipboard
| Challenge: | Existing methods for enhancing the factual accuracy of abstractive summarization models are not effective in fact-critical domains like clinical document summarizing. |
| Approach: | They propose a guided continued pre-training stage for encoder-decoder models followed by supervised fine-tuning on summarization. |
| Outcome: | The proposed approach improves the quality and factuality of the summaries and achieves the best-published results on the clinical document summarization task. |
Benchmarking and Mitigating the Impact of Noisy User Prompts in Medical VLMs via Cross-Modal Reflection (2026.eacl-industry)
Copied to clipboard
| Challenge: | Existing medical vision-language models follow user-provided prompts blindly, a new study finds . current models are noisy, causing problems with reliability in real-world interactions . |
| Approach: | They propose a method to evaluate the influence of clinical prompts on medical vision-language models . they use cross-modal reflection chain-of-thought to train the model to produce reasoning paths . |
| Outcome: | The proposed method significantly improves the robustness against noisy prompts . existing Med-VLMs follow user-provided prompts blindly, the authors show . |
Think Wider, Detect Sharper: Reinforced Reference Coverage for Document-Level Self-Contradiction Detection (2025.emnlp-main)
Copied to clipboard
| Challenge: | Recent approaches to document-level contradiction detection (DSCD) only gain marginal improvement and often introduce inconsistencies across repeated responses. |
| Approach: | They propose a method that combines supervised fine-tuning and reinforcement learning to enhance document-level contradiction detection (DSCD) they propose to use a task-specific reward function to expand the model’s reasoning scope, boosting both accuracy and consistency. |
| Outcome: | The proposed method significantly boosts Llama 3.1-8B-Instruct’s accuracy from 38.5% to 51.1%, and consistency from 59.6% to76.2%. |
Self-Training Large Language Models for Tool-Use Without Demonstrations (2025.findings-naacl)
Copied to clipboard
| Challenge: | Recent work augmented LLMs with tools to mitigate factual inaccuracies and computational errors. |
| Approach: | They propose a method to synthesise tool-use traces using the LLM itself. |
| Outcome: | The proposed method improves performance on a long-tail knowledge task, but not on other datasets. |
Thinking Out Loud: Do Reasoning Models Know When They’re Right? (2025.emnlp-main)
Copied to clipboard
| Challenge: | Large reasoning models (LRMs) have recently demonstrated impressive capabilities in complex reasoning tasks by leveraging increased test-time computation and exhibiting behaviors reminiscent of human-like self-reflection. |
| Approach: | They analyze verbalized confidence, how models articulate their certainty, as a lens into the nature of self-reflection in large reasoning models. |
| Outcome: | The proposed model exhibits human-like self-reflection in reasoning tasks, but how this ability interacts with other model behaviors remains underexplored. |
Ryze: Evidence-Enriched Data Synthesis from Biomedical Papers (2026.acl-demo)
Copied to clipboard
| Challenge: | Existing post-training pipelines that generate QA pairs require costly expert annotation and synthetic data that drops evidence structure. |
| Approach: | They propose a system that converts raw biomedical papers into evidence-enriched training sets and a domain-specialized VLM. |
| Outcome: | Ryze synthesizes QA pairs with complete supporting evidence, reduces layout and OCR errors . the system outperforms the base model on LAB-Bench and surpasses GPT-5.2 by +3.8%. |
Self-Explore: Enhancing Mathematical Reasoning in Language Models with Fine-grained Rewards (2024.findings-emnlp)
Copied to clipboard
| Challenge: | Recent studies have shown that large language models can solve complex reasoning tasks with Chain-of-Thought Prompting. |
| Approach: | They propose a training method where the LLM is tasked to explore the first wrong step within the rationale and use such signals as fine-grained rewards for further improvement. |
| Outcome: | The proposed model improves on the GSM8K and MATH test sets by 11.57% and 2.89% on average compared to supervised fine-tuning (SFT). |
BiLD: Bi-directional Logits Difference Loss for Large Language Model Distillation (2025.coling-main)
Copied to clipboard
| Challenge: | Knowledge distillation (KD) is a method for reducing model size while preserving performance. |
| Approach: | They propose a method to distill large language models at the logit level by transferring knowledge from a large teacher model to a smaller student model. |
| Outcome: | The proposed method outperforms supervised fine-tuning, vanilla KL loss and five other distillation methods on 13 datasets. |
RuleR: Improving LLM Controllability by Rule-based Data Recycling (2025.naacl-short)
Copied to clipboard
| Challenge: | Existing supervised fine-tuning datasets are composed of general instructions without userspecified constraints. |
| Approach: | They propose a data augmentation method incorporating multiple constraints into the original data samples according to predefined rules to create new training tasks. |
| Outcome: | The proposed method improves LLM controllability while maintaining general instruction-following capabilities. |
Training Language Models to Critique With Multi-agent Feedback (2025.findings-emnlp)
Copied to clipboard
Tian Lan, Wenwei Zhang, Chengqi Lyu, Shuaibin Li, Chen Xu, Heyan Huang, Dahua Lin, Xian-Ling Mao, Kai Chen
| Challenge: | utilizing human annotations can enhance critique ability, but model-generated critiques suffer from inherent flaws due to complexity of critique . a new framework that leverages multi-agent feedback improves critique ability . |
| Approach: | They propose a framework that leverages multi-agent feedback to improve critique ability . they propose to use supervised fine-tuning and reinforcement learning to improve this capability . |
| Outcome: | The proposed framework improves critique ability in both supervised fine-tuning and reinforcement learning stages. |
Hierarchical Policy Optimization for Simultaneous Translation of Unbounded Speech (2026.acl-long)
Copied to clipboard
| Challenge: | Existing synthesis methods cannot guarantee data quality. |
| Approach: | They propose a hierarchical reward that balances translation quality and latency objectives by combining supervised fine-tuning data with supervised inputs. |
| Outcome: | The proposed model can reuse key-value caches across both modalities and eliminate redundant feature recomputation. |
Towards Tool Use Alignment of Large Language Models (2024.emnlp-main)
Copied to clipboard
| Challenge: | Existing studies on tool use with LLMs focus on enhancing tool-calling ability of LLM . e.g., LLM should not answer unsafe tool use relevant instructions or insecure tool responses to ensure reliability and harmlessness. |
| Approach: | They propose to use supervised fine-tuning and preference learning to align LLMs with H2A principle for tool use. |
| Outcome: | The proposed model demonstrates that LLMs can generate truthful and helpful responses while remaining harmless. |
RECIPE-TKG: From Sparse History to Structured Reasoning for LLM-based Temporal Knowledge Graph Completion (2026.eacl-long)
Copied to clipboard
| Challenge: | Temporal Knowledge Graphs (TKGs) represent dynamic facts as timestamped relations between entities. Large Language Models (LLMs) have sparked interest in using pretrained generative models for TKG completion. |
| Approach: | They propose a framework that allows for rule-based multi-hop sampling and contrastive fine-tuning to shape relational compatibility. |
| Outcome: | Experiments show that RECIPE-TKG outperforms prior LLM-based methods across input regimes. |
RedOne 2.0: Rethinking Domain-specific LLM Post-Training in Social Networking Services (2026.acl-industry)
Copied to clipboard
Fei zhao, Chonggang Lu, Haofu Qian, Fangcheng Shi, Zijie Meng, Jianzhao Huang, Zheyong Xie, Shaosheng Cao
| Challenge: | Social networking services (SNS) are critical infrastructure for global interaction . supervised fine-tuning (SFT) can improve in-domain performance, but it often induces a ”seesaw” trade-off with out-of-domain robustness . |
| Approach: | They propose an SNS-oriented LLM with a progressive, RL-prioritized post-training paradigm for fast and stable adaptation. |
| Outcome: | The proposed model improves over the previous 7B model by 2.41 on average . it also yields an 8.74 average gain over its Qwen3-4B base . |
We Politely Insist: Your LLM Must Learn the Persian Art of Taarof (2025.emnlp-main)
Copied to clipboard
| Challenge: | a new study examines the cultural competence of large language models to navigate complex social interactions. |
| Approach: | They propose a benchmark to evaluate LLM understanding of Persian taarof . they evaluate 450 role-play scenarios covering 12 common social interaction topics . their results show significant gaps in cultural competence . |
| Outcome: | The proposed model improves on Persian *taarof*, a social norm in Iranian interactions . the model improve despite culturally specific responses and gender-based asymmetries . |
SimRPD: Optimizing Recruitment Proactive Dialogue Agents through Simulator-Based Data Evaluation and Selection (2026.acl-industry)
Copied to clipboard
Zhiyong Cao, Dunqiang Liu, Qi Dai, Haojun Xu, Huai Yuen Khor, Hao Wang, Huan He, Yafei Liu, Ke Ma, Ruqian Shi, Sicheng Zhou, Sijia Yao
| Challenge: | High-quality data in training proactive dialogue agents is scarce, despite fine-tuning and reinforcement learning . a recent study has shown that the effectiveness of supervised fine-touring is limited by the lack of high-quality, domain-specific training data. |
| Approach: | They propose a framework for training recruitment proactive dialogue agents using a high-fidelity user simulator and a multi-dimensional evaluation framework based on Chain-of-Intention. |
| Outcome: | The proposed framework outperforms existing simulator-based data selection strategies in a real-world recruitment scenario. |
Developing and Utilizing a Large-Scale Cantonese Dataset for Multi-Tasking in Large Language Models (2025.findings-emnlp)
Copied to clipboard
Jiyue Jiang, Alfred Kar Yin Truong, Yanyu Chen, Qinghang Bao, Sheng Wang, Pengan Chen, Jiuming Wang, Lingpeng Kong, Yu Li, Chuan Wu
| Challenge: | Cantonese is considered a low-resource language due to the dominance of Mandarin . rich colloquial vocabulary of Cantone, English loanwords, and code-switching characteristics add to the complexity of corpus collection and processing. |
| Approach: | We collect Cantonese texts from open source corpora, Hong Kong-specific forums, Wikipedia . we refine the model through supervised fine-tuning on curated Cantonesian tasks . |
| Outcome: | The model achieves state-of-the-art (SOTA) performance on four Cantonese benchmarks. |
Solver-Independent Automated Problem Formulation via LLMs for High-Cost Simulation-Driven Design (2026.findings-acl)
Copied to clipboard
| Challenge: | Existing approaches to translating ambiguous design requirements into a mathematical optimization formulation are expensive and time-consuming. |
| Approach: | They propose a solver-independent framework that converts engineers’ natural language requirements into executable optimization models. |
| Outcome: | The proposed framework outperforms existing methods in the accuracy of requirement formalization and quality of resulting radiation efficiency curves on antenna design. |
Train a Unified Multimodal Data Quality Classifier with Synthetic Data (2025.findings-emnlp)
Copied to clipboard
Weizhi Wang, Rongmei Lin, Shiyang Li, Colin Lockard, Ritesh Sarkhel, Sanket Lokegaonkar, Jingbo Shang, Xifeng Yan, Nasser Zalmout, Xian Li
| Challenge: | Multimodal Large Language Models are pre-trained on image-text caption data and interleaved document data. |
| Approach: | They propose to train an efficient MLLM as a Unified Mulitmodal Data Quality Classifier to filter image-text caption and interleaved data. |
| Outcome: | The proposed method enables efficient creation of sample-score pairs for caption and interleaved data to train UniFilter. |
Knowledge-Augmented Question Error Correction for Chinese Question Answer System with QuestionRAG (2025.emnlp-industry)
Copied to clipboard
| Challenge: | Large language models struggle with input errors, often failing to interpret user intent or altering the original question’s structure (over-correction). |
| Approach: | They propose a framework that uses reinforcement learning to address misinterpretation and over-correction by integrating external knowledge with the input. |
| Outcome: | The proposed framework unlocks the full potential of LLMs for the question correction task. |
Think-Search-Patch: A Retrieval-Augmented Reasoning Framework for Repository-Level Code Repair (2025.emnlp-industry)
Copied to clipboard
Bojian Xiong, Yikun Lei, Xikai Liu, Shaowei Zhang, Pengyun Zhu, Yan Liu, Yongqi Leng, Ling Shi, Meizhi Zhong, Yurong Zhang, Yan Gao, null Yiwu, Yao Hu, Deyi Xiong
| Challenge: | Large language models suffer from multiple-file coding scenarios with strong inter-file dependencies . experimental results show that large language models exhibit inadequate performance in multi-file scenarios . |
| Approach: | They propose a retrieval-augmented reasoning framework for repository-level code repair . they use a dataset to generate standardized patches based on the key snippets . |
| Outcome: | The proposed framework improves retrieval accuracy and repair success on SWE-bench Lite . it surpasses models with larger size in managing extensive code contexts and fixing bugs spanning across multiple files. |
SDE-SQL: Enhancing Text-to-SQL Generation in Large Language Models via Self-Driven Exploration with SQL Probes (2026.acl-long)
Copied to clipboard
| Challenge: | Existing approaches depend on static, pre-processed database information, which restricts the model’s capacity to deeply comprehend the underlying database content. |
| Approach: | They propose a framework that empowers LLMs to perform Self-Driven Exploration of databases during inference. |
| Outcome: | Evaluated on the BIRD benchmark with Qwen2.5-72B-Instruct, SDE-SQL achieves an 8.02 % improvement in execution accuracy over the baseline. |
LegalDrill: Diagnosis-Driven Synthesis for Legal Reasoning in Small Language Models (2026.acl-industry)
Copied to clipboard
Tianchun Li, Haochen Liu, Vishwa Pardeshi, Xingchen Wang, Tianci Liu, Huijun Zhao, Wei Fan, Jing Gao
| Challenge: | Small language models (SLMs) are promising for real-world deployment but struggle with high-stakes legal reasoning tasks. |
| Approach: | They propose a diagnostic-driven synthesis framework that extracts and refines reasoning trajectories from a capable teacher via fine-grained prompting and a self-reflective verification is employed to adaptively select the most effective data for the SLM student. |
| Outcome: | The proposed framework extracts and refines reasoning trajectories from a capable teacher via fine-grained prompting, then a self-reflective verification is employed to adaptively select the most effective data for the student. |
MPL: Multiple Programming Languages with Large Language Models for Information Extraction (2025.findings-acl)
Copied to clipboard
| Challenge: | Existing research focuses on Python for code-style simulation, overlooking the potential of other widely-used PLs during the supervised fine-tuning phase. |
| Approach: | They propose a framework that incorporates programming languages into IE tasks . they introduce function-prompt with virtual running to simulate code-style inputs . |
| Outcome: | The proposed framework exploits the potential of different programming languages during the supervised fine-tuning phase. |
LightReasoner: Can Small Language Models Teach Large Language Models Reasoning? (2026.acl-long)
Copied to clipboard
| Challenge: | Large language models (LLMs) have demonstrated remarkable progress in reasoning, but are resource-intensive and require large curated datasets. |
| Approach: | They propose a framework that leverages the behavioral divergence between a stronger expert model and a weaker amateur model. |
| Outcome: | The proposed framework improves accuracy by up to 28.1% while reducing time consumption by 90% and tuning token usage by 99%. |
Beyond Surface-Level Detection: Towards Cognitive-Driven Defense Against Jailbreak Attacks via Meta-Operations Reasoning (2026.acl-long)
Copied to clipboard
| Challenge: | Existing defenses rely on shallow pattern matching, which struggles to generalize to novel and unseen attack strategies. |
| Approach: | They propose a framework which emulates human cognitive reasoning through a structured reasoning chain. |
| Outcome: | The proposed framework achieves state-of-the-art performance and exhibits strong generalization to unseen attacks. |
ALIGN: Word Association Learning for Cultural Alignment in Large Language Models (2026.acl-long)
Copied to clipboard
| Challenge: | Large language models exhibit cultural bias from over-represented viewpoints in training data, yet cultural alignment remains a challenge due to limited cultural knowledge and a lack of exploration into effective learning approaches. |
| Approach: | They propose a cost-efficient method for fine-tuning large language models on native speakers’ word-association norms and a preference optimization method to improve cultural alignment. |
| Outcome: | The proposed model trains Llama-3.1-8B and Qwen-2.5-7B on native speakers’ word-association norms and shows that such associations capture cultural knowledge. |
Eliciting Medical Reasoning with Knowledge-enhanced Data Synthesis: A Semi-Supervised Reinforcement Learning Approach (2026.findings-acl)
Copied to clipboard
| Challenge: | Existing methods to enhance medical reasoning lack high-quality data. |
| Approach: | They propose a medical knowledge-enhanced data Synthesis and Semi-supervised Reinforcement learning framework that uses rare disease knowledge to synthesize distribution-controllable reasoning questions. |
| Outcome: | The proposed method outperforms existing methods across ten medical benchmarks and achieves up to 5.93% gain on rare diseases tasks. |
CriticLean: Critic-Guided Reinforcement Learning for Mathematical Formalization (2026.acl-long)
Copied to clipboard
Zhongyuan Peng, Yifan Yao, Kaijing Ma, Shuyue Guo, Yizhe Li, Yichi Zhang, Chenchen Zhang, Yifan Zhang, Zhouliang Yu, Luming Li, Minghao Liu, Yihang Xia, Jiawei Shen, Yuchen Wu, Yixin Cao, Zhaoxiang Zhang, Wenhao Huang, Jiaheng Liu, Ge Zhang
| Challenge: | Existing approaches to formalizing mathematical statements face limitations in accuracy, especially in the context of complex, highlevel problems that involve sophisticated mathematical reasoning. |
| Approach: | They propose a CriticLean framework that elevates the role of the critic from a passive validator to an active learning component and introduce a benchmark to measure models’ ability to distinguish semantically correct from incorrect formalizations. |
| Outcome: | The proposed framework outperforms open- and closed-source benchmarks and shows that it significantly outperformed existing models. |
Self-Refine Instruction-Tuning for Aligning Reasoning in Language Models (2024.emnlp-main)
Copied to clipboard
| Challenge: | Existing approaches to align reasoning abilities between Large Language Models and Smaller Language Model are supervised fine-tuning and preference optimization. |
| Approach: | They propose a method that elicits Smaller Language Models to self-improve their reasoning abilities via preference optimization. |
| Outcome: | The proposed method outperforms Instruction-tuning on commonsense and math reasoning tasks on common and math scenarios. |
AVATAR: A Parallel Corpus for Java-Python Program Translation (2023.findings-acl)
Copied to clipboard
| Challenge: | Program translation is a time-consuming and costly process that requires expertise in both the source and target languages. |
| Approach: | They present a collection of 9,515 programming problems and their solutions written in Java and Python. |
| Outcome: | The proposed model lacks in generating functionally accurate code. |
LSDC: An Efficient and Effective Large-Scale Data Compression Method for Supervised Fine-tuning of Large Language Models (2025.findings-naacl)
Copied to clipboard
Zhaoguang Long, Yuhao Zhou, Shangqing Zhao, Yupei Ren, Li Cai, Chenghao Jia, Zhe Chen, Zhe Fang, Yuxiang Song, Man Lan
| Challenge: | Large Language Models (LLMs) are expanding in scale and size, increasing computational costs . large-scale data compression techniques can reduce the size of training datasets while maintaining data integrity. |
| Approach: | They propose a large-scale data compression method to reduce the size of training data . they use a bifurcated quantization strategy to maximize the diversity of samples . |
| Outcome: | The proposed method significantly reduces the size of training data while maximizing the submodular gain. |
CrossQG: Improving Difficulty-Controllable Question Generation through Consistency Enhancement (2025.findings-emnlp)
Copied to clipboard
| Challenge: | Large language models can generate questions with controlled difficulty, but they often fail to align with the given target difficulty. |
| Approach: | They propose a question generation method that requires no tuning of generator parameters yet significantly improves difficulty consistency. |
| Outcome: | The proposed method outperforms several mainstream methods on high-quality question answering datasets and achieves superior consistency with target difficulty. |
MIThinker: A Plug-and-Play Policy-Optimized Thinker For Motivational Interviewing Counseling (2026.findings-acl)
Copied to clipboard
| Challenge: | Existing reasoning large language models (LLMs) generate responses without explicitly aligning thoughts with counseling techniques, limiting their effectiveness. |
| Approach: | They propose a lightweight thinking model that generates therapeutic thoughts to guide MI counseling agents in strategy selection and response generation. |
| Outcome: | The proposed model achieves theory-of-mind assessment comparable to state-of the-art systems with an order of magnitude less computation. |
Self-Renewal Prompt Optimizing with Implicit Reasoning (2024.findings-emnlp)
Copied to clipboard
Zihan Liang, Ben Chen, Zhuoran Ran, Zihan Wang, Huangyu Dai, Yufei Ma, Dehong Gao, Xiaoyan Cai, Libin Yang
| Challenge: | Recent advances in NLP have been driven by the development of Large Language Models (LLMs). |
| Approach: | They propose a self-renewal approach to optimize LLM outputs to better align with human preferences without supervised fine-tuning. |
| Outcome: | The proposed approach improves outputs to better align with human preferences across LLMs and tasks without supervised fine-tuning. |
Domaino1s: Guiding LLM Reasoning for Explainable Answers in High-Stakes Domains (2025.findings-acl)
Copied to clipboard
| Challenge: | Existing Large Language Models (LLMs) generate brief answers without reasoning processes and explanations. |
| Approach: | They propose supervised fine-tuning and tree search to enhance LLMs’ reasoning capabilities on domain tasks. |
| Outcome: | The proposed model improves on stock investment recommendation and legal reasoning QA tasks. |
Anchoring-Guidance Fine-Tuning (AnGFT): Elevating Professional Response Quality in Role-Playing Conversational Agents (2025.emnlp-main)
Copied to clipboard
| Challenge: | Large Language Models (LLMs) have demonstrated significant advancements in various fields, notably in Role-Playing Conversational Agents (RPCAs). |
| Approach: | They propose an Anchoring-Guidance Fine-Tuning Framework to integrate relevant expert knowledge into RPCAs' training process to mitigate this issue. |
| Outcome: | The proposed framework significantly improves the RPCAs’ performance in handling role-specific professional queries while preserving their robust role-playing abilities. |
DRIVINGVQA: A Dataset for Interleaved Visual Chain-of-Thought in Real-World Driving Scenarios (2026.findings-eacl)
Copied to clipboard
| Challenge: | Chain-of-thought (CoT) prompting is a prompting strategy that improves reasoning in large language models, but its effectiveness in vision-language models remains limited due to over-reliance on textual cues and memorized knowledge. |
| Approach: | They propose a visual question-answering dataset derived from driving theory exams that incorporates textual explanations with visual tokens extracted from entities relevant to the reasoning process. |
| Outcome: | The proposed approach outperforms chain-of-thought prompting in large language models and vision-language models in real-world scenarios. |
RedOne: Revealing Domain-specific LLM Post-Training in Social Networking Services (2025.emnlp-industry)
Copied to clipboard
Fei Zhao, Chonggang Lu, null Wangyue, Zheyong Xie, Ziyan Liu, Haofu Qian, Jianzhao Huang, Fangcheng Shi, Zijie Meng, Hongcheng Guo, Mingqian He, Xinze Lyu, Zheyu Ye, Weiting Liu, Boyang Wang, Shaosheng Cao
| Challenge: | Social networking services (SNS) have experienced rapid growth, which has proposed significant challenges for platform content management and interaction quality improvement. |
| Approach: | They propose a domain-specific LLM to break the performance bottleneck of single-task baselines and establish a comprehensive foundation for social networking services. |
| Outcome: | The proposed model achieves an average improvement of 14.02% across 8 major tasks and 7.56% in bilingual evaluation benchmark, compared with baseline models. |
Enhancing the General Agent Capabilities of Low-Paramter LLMs through Tuning and Multi-Branch Reasoning (2024.findings-naacl)
Copied to clipboard
| Challenge: | Open-source pre-trained Large Language Models exhibit strong language understanding and generation capabilities, making them highly successful in a variety of tasks. |
| Approach: | They propose a method to construct agent-specific data using GPT-4 and supervised fine-tuning . they find that supervised tunning can significantly reduce hallucination outputs and formatting errors in agent tasks . |
| Outcome: | The proposed method improves on five agent tasks of AgentBench. |
LitE-SQL: A Lightweight and Efficient Text-to-SQL Framework with Vector-based Schema Linking and Execution-Guided Self-Correction (2026.findings-eacl)
Copied to clipboard
| Challenge: | Existing methods rely on proprietary models to generate SQL queries. |
| Approach: | They propose a lightweight framework that translates natural language questions into SQL queries. |
| Outcome: | The proposed framework achieves 72.10% execution accuracy on BIRD and 88.45% on Spider 1.0 . it offers a practical solution for privacy-sensitive and resource-constrained settings. |
A Preference-driven Paradigm for Enhanced Translation with Large Language Models (2024.naacl-long)
Copied to clipboard
| Challenge: | Recent research shows that large language models (LLMs) can achieve remarkable translation performance through supervised fine-tuning (SFT) however, SFT simply instructs the model to imitate reference translations token by token, making it vulnerable to the noise present in the data. |
| Approach: | They propose a preference-based approach to supervised fine-tuning that trains the model to imitate reference translations token by token, making it vulnerable to noise. |
| Outcome: | The proposed approach overcomes the plateau associated with imitation-based SFT and is more resilient in the absence of gold translations. |
LongReward: Improving Long-context Large Language Models with AI Feedback (2025.acl-long)
Copied to clipboard
Jiajie Zhang, Zhongni Hou, Xin Lv, Shulin Cao, Zhenyu Hou, Yilin Niu, Lei Hou, Yuxiao Dong, Ling Feng, Juanzi Li
| Challenge: | In recent years, significant advancements have been achieved in the development of long-context large language models (LLMs). |
| Approach: | They propose a method that utilizes an off-the-shelf LLM to provide rewards for long-context model responses from four human-valued dimensions: helpfulness, logicality, faithfulness, and completeness. |
| Outcome: | The proposed method improves models’ long-context performance and enhances their ability to follow short instructions. |
Learning SQL Like a Human: Structure-Aware Curriculum Learning for Text-to-SQL Generation (2025.findings-emnlp)
Copied to clipboard
| Challenge: | Existing models struggle with complex queries, especially multi-table joins and reasoning. |
| Approach: | They propose to build a model with synthetic training samples and a structure-aware curriculum learning framework for enhancing SQL generation. |
| Outcome: | The proposed model improves on the existing model on the Spider and Bird benchmarks. |
Making Language Models Better Tool Learners with Execution Feedback (2024.naacl-long)
Copied to clipboard
| Challenge: | Existing tool learning methodologies induce large language models to utilize tools indiscriminately . Existing frameworks that teach language models when and how to use tools propagate errors rather than enhance performance. |
| Approach: | They propose a framework that enables large language models to continually learn through feedback derived from tool execution. |
| Outcome: | The proposed framework can make large language models selectively use tools . it improves accuracy while enhancing insufficient tool learning, it shows . |
BitBypass: A New Direction in Jailbreaking Aligned Large Language Models with Bitstream Camouflage (2026.findings-eacl)
Copied to clipboard
| Challenge: | Large Language Models (LLMs) have inherent risk of generating harmful and unsafe content. |
| Approach: | They develop a black-box jailbreak attack that leverages hyphen-separated bitstream camouflage to bypass aligned Large Language Models' safety alignment. |
| Outcome: | The proposed attack outperforms state-of-the-art jailbreak attacks in stealthiness and attack success. |
AceMath: Advancing Frontier Math Reasoning with Post-Training and Reward Modeling (2025.findings-acl)
Copied to clipboard
| Challenge: | In this paper, we introduce a suite of math models that excel in solving complex math problems. |
| Approach: | They propose a supervised fine-tuning process that achieves competitive performance across general domains, followed by targeted fine- tuning for the math domain using a carefully curated set of prompts and synthetically generated responses. |
| Outcome: | The proposed model outperforms Qwen2.5-Math-72B-Instruct, GPT-4o and Claude-3.5 Sonnet in the math domain. |
Timely Machine: Awareness of Time Makes Test-Time Scaling Agentic (2026.acl-long)
Copied to clipboard
| Challenge: | Large language models (LLMs) tackle complex reasoning tasks, but test-time scaling is becoming expensive. |
| Approach: | They propose to redefine test-time as wall-clock time, where models dynamically adjust strategies based on time budgets. |
| Outcome: | The proposed model improves time budget awareness and boosts performance across Timely-Eval. |
D2Plan: Dual-Agent Dynamic Global Planning for Complex Retrieval-Augmented Reasoning (2026.acl-long)
Copied to clipboard
| Challenge: | Recent advances in reinforcement learning (RL) have empowered Large Language Models (LLMs) with the capability to perform autonomous retrieval during reasoning tasks. |
| Approach: | They propose a "D2Plan" paradigm for retrieval-augmented reasoning that integrates a 'Reasoner' and a'Purifier' |
| Outcome: | Experiments show that the proposed paradigm improves on QA benchmarks. |
COF: Adaptive Chain of Feedback for Comparative Opinion Quintuple Extraction (2025.coling-main)
Copied to clipboard
| Challenge: | Comparative Opinion Quintuple Extraction (COQE) aims to extract all comparative sentiment quintuples from product review text. |
| Approach: | They propose a model-unaware adaptive chain-of-feedback method to extract quintuples from product review text. |
| Outcome: | The proposed method improves performance on three benchmarks. |
Towards Order Fairness: Mitigating LLMs Order Sensitivity through Dual Group Advantage Optimization (2026.acl-long)
Copied to clipboard
| Challenge: | Recent studies attempt to obtain optimal or suboptimal arrangements based on statistical results or using dataset-based search, but these methods increase inference overhead while leaving the model’s inherent order bias unresolved. |
| Approach: | They propose Dual Group Advantage Optimization (DGAO) which aims to improve model accuracy and order stability simultaneously. |
| Outcome: | The proposed method improves model accuracy and order stability while penalizing order-sensitive or incorrect responses. |
Watermarking LLMs with Weight Quantization (2023.findings-emnlp)
Copied to clipboard
| Challenge: | Large language models are being deployed at an astonishing speed, exposing users to high risks. |
| Approach: | They propose a method that plants watermarks in quantization process of large language models without pre-defined triggers during inference. |
| Outcome: | The proposed method protects model weights without pre-defined triggers . it works when the model is used in the fp32 mode and remains hidden when the models are quantized to int8 . |
Uncovering Temporal Framing in the News (2026.acl-long)
Copied to clipboard
Tarek Mahmoud, Veronika Solopova, Premtim Sahitaj, Ariana Sahitaj, Max Upravitelev, Mervat Abassy, Hana Fatima Shaikh, Neda Foroutan, Vera Schmitt, Preslav Nakov
| Challenge: | Temporal language is used to structure meaning rather than report chronology in news discourse . a recent study focused on temporal expression extraction and temporal reasoning . |
| Approach: | They propose a taxonomy of eight temporal frames grounded in prior work on time and framing . they analyze frame prevalence, co-occurrence patterns, and lexical cues from a news corpus . |
| Outcome: | The proposed taxonomy outperforms zero-shot models at the sentence level . it shows that temporal framing is learnable at the sentences level compared to other methods . |
David vs. Goliath: Cost-Efficient Financial QA via Cascaded Multi-Agent Reasoning (2025.findings-emnlp)
Copied to clipboard
| Challenge: | Large language models have demonstrated remarkable reasoning capabilities, but performance in FQA remains limited. |
| Approach: | They propose a low-cost yet effective framework that enables small LLMs to perform complex reasoning tasks without expensive models. |
| Outcome: | The proposed framework outperforms the best open-source model on BizBench by 10.46% and achieves competitive performance to GPT-3.5 using significantly fewer parameters. |
Self-Evolution Fine-Tuning for Policy Optimization (2024.findings-emnlp)
Copied to clipboard
| Challenge: | Recent years have showcased the remarkable capabilities and performance of large language models (LLMs) across a broad range of tasks. |
| Approach: | They propose supervised fine-tuning (SEFT) for LLM alignment to eliminate the need for annotated samples while retaining the stability and efficiency of SFT. |
| Outcome: | The proposed method eliminates the need for annotated samples while maintaining the stability and efficiency of SFT. |
Reasoning Structure Matters for Safety Alignment of Reasoning Models (2026.acl-long)
Copied to clipboard
| Challenge: | Large reasoning models (LRMs) achieve strong performance on complex reasoning tasks but often generate harmful responses to malicious user queries. |
| Approach: | They propose a method that alters the reasoning structure of large reasoning models to achieve effective safety alignment by supervised fine-tuning. |
| Outcome: | The proposed method is practical and generalizable, requiring no complex training or reward design. |
Backdoor-Powered Prompt Injection Attacks Nullify Defense Methods (2025.findings-emnlp)
Copied to clipboard
| Challenge: | Recent studies have shown that LLMs are vulnerable to prompt injection attacks because of their instruction-following abilities and inability to distinguish the instructions in the data content. |
| Approach: | They propose backdoor-powered prompt injection attacks that trick LLMs into deviating from the original input instruction and executing the attackers’ target instruction. |
| Outcome: | The proposed attacks trick the LLMs into deviating from the input instruction and executing the attackers’ target instruction. |
When Audio and Text Disagree: Revealing Text Bias in Large Audio-Language Models (2025.emnlp-main)
Copied to clipboard
| Challenge: | Large Audio-Language Models (LALMs) are augmented with the ability to perceive audio, but their reliability when faced with conflicting inputs remains largely unexplored. |
| Approach: | They examine how LALMs prioritize information when presented with inconsistent audio-text pairs. |
| Outcome: | The proposed models display a significant bias toward textual input when presented with inconsistent audio-text pairs. |
Cross-Cultural Transfer of Commonsense Reasoning in LLMs: Evidence from the Arab World (2025.findings-emnlp)
Copied to clipboard
| Challenge: | a recent study examined the potential for cross-cultural transfer of commonsense reasoning . merely 12 culture-specific examples from one country can improve performance in others by 10% on average . |
| Approach: | They evaluate cross-cultural transfer of commonsense reasoning within the arab world . they use in-context learning and demonstration-based reinforcement to evaluate alignment methods . |
| Outcome: | The proposed model can improve performance in cultures with cultural similarities in the Arab world by 10% on average. |
Prejudge-Before-Think: Enhancing Large Language Models at Test-Time by Process Prejudge Reasoning (2025.findings-emnlp)
Copied to clipboard
| Challenge: | Experimental results from competition-level complex reasoning demonstrate that bootstrapping with process prejudge can significantly enhance the reasoning ability of LLMs. |
| Approach: | They propose a new process prejudge strategy for LLM reasoning that bootstraps with process prejudgment . |
| Outcome: | The proposed method can be bootstrapped with process prejudge in LLM reasoning . it allows the model to anticipate errors rather than relying on trial and error. |
Through the Valley: Path to Effective Long CoT Training for Small Language Models (2025.emnlp-main)
Copied to clipboard
| Challenge: | Long chain-of-thought (CoT) supervision is effective for large language models . but small models trained on limited long CoT data experience performance degradation . |
| Approach: | They identify a phenomenon called Long CoT Degradation in small language models . long CoT data can be used to generate long chain-of-thought (CoT) responses . |
| Outcome: | The results show that models trained on 8k long CoT examples lose up to 75% of their original performance before fine-tuning. |
Zero-to-Strong Generalization: Eliciting Strong Capabilities of Large Language Models Iteratively without Gold Labels (2025.coling-main)
Copied to clipboard
| Challenge: | Pre-trained language models have demonstrated remarkable performance through supervised fine-tuning or in-context learning using gold labels. |
| Approach: | They propose a new paradigm termed zero-to-strong generalization that prompts LLMs to annotate unlabeled data and retain high-quality labels by filtering. |
| Outcome: | The proposed framework outperforms pre-trained language models on extensive classification and reasoning tasks on multiple model sizes. |
Crowd Comparative Reasoning: Unlocking Comprehensive Evaluations for LLM-as-a-Judge (2025.acl-long)
Copied to clipboard
Qiyuan Zhang, Yufei Wang, Yuxin Jiang, Liangyou Li, Chuhan Wu, Yasheng Wang, Xin Jiang, Lifeng Shang, Ruiming Tang, Fuyuan Lyu, Chen Ma
| Challenge: | Existing methods rely on majority voting or criteria expansion to capture detailed and detailed details, often leading to incomplete outcomes. |
| Approach: | They propose a method which introduces additional crowd responses to compare with the candidate responses, thereby exposing deeper and more comprehensive details within the candidate answers. |
| Outcome: | Experiments show that the proposed method improves evaluation reliability and achieves an average gain of 6.7% across five benchmarks. |
MIO: A Foundation Model on Multimodal Tokens (2025.emnlp-main)
Copied to clipboard
Zekun Moore Wang, King Zhu, Chunpu Xu, Wangchunshu Zhou, Jiaheng Liu, Yibo Zhang, Jessie Wang, Ning Shi, Siyu Li, Yizhi Li, Haoran Que, Zhaoxiang Zhang, Yuanxing Zhang, Ge Zhang, Ke Xu, Jie Fu, Wenhao Huang
| Challenge: | Existing models lack multimodal understanding capabilities, resulting in closed-source model that does not support multimodal interleaved sequences. |
| Approach: | They propose a foundation model built on multimodal tokens capable of understanding and generating speech, text, images, and videos in an end-to-end, autoregressive manner. |
| Outcome: | The proposed model is able to understand speech, text, images, and videos in an end-to-end, autoregressive manner. |
Packing Analysis: Packing Is More Appropriate for Large Models or Datasets in Supervised Fine-tuning (2025.findings-acl)
Copied to clipboard
| Challenge: | Packing is an optimization technique that optimizes training time and resources by combining different training sequences to fit the model’s maximum input length. |
| Approach: | They perform extensive comparisons between packing and padding methods, covering datasets ranging from 69K to 1.2M and models from 8B to 70B. |
| Outcome: | The proposed method has been shown to improve training efficiency while maintaining performance. |
Benchmarking Direct Preference Optimization for Medical Large Vision–Language Models (2026.findings-eacl)
Copied to clipboard
| Challenge: | Large vision-language models (LVLMs) are gaining traction in clinical tasks such as diagnostic support, report generation, and medical question answering. |
| Approach: | They present a systematic evaluation of nine DPO variants applied to two leading medical LVLMs. |
| Outcome: | The proposed model improves alignment and reduces severe hallucinations, but yields inconsistent gains over supervised fine-tuning. |
StableToolBench-MirrorAPI: Modeling Tool Environments as Mirrors of 7,000+ Real-World APIs (2025.findings-acl)
Copied to clipboard
| Challenge: | Existing tool environments face challenges in balancing stability, scale, and realism, especially for benchmarking purposes. |
| Approach: | They propose a framework that trains specialized LLMs to accurately simulate real API responses by supervised fine-tuning and chain-of-thought reasoning. |
| Outcome: | The proposed framework achieves superior accuracy and stability compared to state-of-the-art methods on the newly constructed MirrorAPI-Bench and its integration into StableToolBench. |
ORSO QGen: Odds-Ratio Steerable Optimization for Controlling Question Generation (2026.findings-eacl)
Copied to clipboard
| Challenge: | Existing methods for question generation rely on supervised fine-tuning . scarcity of datasets with multiple fine-grained attributes limits ability of models to generalize or transfer control to new combinations of attributes. |
| Approach: | They propose a framework to enhance attribute sensitivity in question generation models . they propose supervised fine-tuning with explicit attribute labels to produce questions that fit predefined characteristics. |
| Outcome: | The proposed framework improves attribute sensitivity while maintaining quality of output. |
Narrative Style and the Spread of Health Misinformation on Twitter (2023.findings-emnlp)
Copied to clipboard
| Challenge: | Using a narrative style is an effective way to communicate health information on and off social media platforms. |
| Approach: | They annotate health misinformation tweets and classify them into narrative and non-narrative . they then use supervised fine-tuning and in-context learning to detect narratives . |
| Outcome: | The proposed model analyzes health misinformation tweets and finds that narrative use is linked to increased tweet engagement and can lead to increased misinformation use. |
From the Least to the Most: Building a Plug-and-Play Visual Reasoner via Data Synthesis (2024.emnlp-main)
Copied to clipboard
| Challenge: | Existing models consisting of multiple steps of visual and language processing are limited in the visual and visual processing community . a visual reasoner is a plug-and-play approach that can be used to improve VLMs' reasoning abilities. |
| Approach: | They propose a least-to-most visual reasoning paradigm that divides a question into sub-questions and invokes external tools for resolving sub-questions. |
| Outcome: | The proposed method can improve four VLMs on four VQA benchmarks. |
Revealing the Inherent Instructability of Pre-Trained Language Models (2025.findings-emnlp)
Copied to clipboard
| Challenge: | Pre-trained large language models perform multitask learning during their pre-training . a new technique, Response Tuning, removes the instruction and its corresponding mapping to the response from instruction tuning. |
| Approach: | They propose a method which removes the instruction and its mapping to the response from instruction tuning. |
| Outcome: | The proposed model can respond to a wide range of instructions . it can recognize and reject unsafe queries after learning from response data. |
From "Aha Moments" to Controllable Thinking: Toward Meta-Cognitive Reasoning in LRMs via Decoupled Reasoning and Control (2026.acl-long)
Copied to clipboard
| Challenge: | Large Reasoning Models exhibit step-by-step reasoning, reflection, and backtracking, but these behaviors are often unregulated, leading to overthinking. |
| Approach: | They propose a meta-cognitive reasoning framework that decouples reasoning from control to enable independent optimization of control strategies. |
| Outcome: | Experiments show that the proposed model improves efficiency and accuracy across reasoning benchmarks. |
GOBench: Stage-Wise Diagnostics and the Visual Paradox in Multimodal Graph Optimization (2026.findings-acl)
Copied to clipboard
| Challenge: | Existing benchmarks fail to represent multimodal problem specifications, score outcomes only and cannot localize where failures occur along the modeling pipeline. |
| Approach: | They propose a Graph Optimization benchmark that aligns multiple modalities with solver-derived oracles and a diagnostic protocol that evaluates intermediate artifacts as well as end results. |
| Outcome: | Graph Optimization benchmark (GOBench) evaluates intermediate artifacts as well as end results . vision reliably increases inference cost, while reliability impact is regime-dependent . current benchmarks fail to represent multimodal problem specifications, fail to localize failures . |
Diff4TST: Masked Diffusion Language Model for Text Style Transfer (2026.acl-long)
Copied to clipboard
| Challenge: | Existing methods for text style transfer rely on task-specific training and expensive training stages. |
| Approach: | They propose a diffusion-based language model that formulates text style transfer as an explicit copy-and-edit process. |
| Outcome: | The proposed model improves style accuracy and controllability while maintaining strong content preservation and fluency. |
Calibrating LLMs with Preference Optimization on Thought Trees for Generating Rationale in Science Question Scoring (2024.findings-emnlp)
Copied to clipboard
| Challenge: | Existing methods for generating rationales that justify scoring decisions are not accurate and often contain hallucinated information. |
| Approach: | They propose a framework capable of generating more faithful rationales and matching performance with classifier-based scoring systems. |
| Outcome: | The proposed framework achieves 38% improvement in QWK score compared to prior work . it can be used to match performance with classifier-based scoring systems . |
ELAINE-medLLM: Lightweight English Japanese Chinese Trilingual Large Language Model for Bio-medical Domain (2025.coling-main)
Copied to clipboard
Ken Yano, Zheheng Luo, Jimin Huang, Qianqian Xie, Masaki Asada, Chenhan Yuan, Kailai Yang, Makoto Miwa, Sophia Ananiadou, Jun’ichi Tsujii
| Challenge: | Existing bilingual or multilingual medical LLMs are limited in multilingual data and therefore perform poorly in non-English languages such as Japanese and Chinese. |
| Approach: | They propose to use a trilingual (English, Japanese, Chinese) large language model adapted for the bio-medical domain to harness the knowledge and abilities of the base model. |
| Outcome: | The proposed model can support English, Japanese, and Chinese and is adapted for a bio-medical domain. |
Aligning Text, Code, and Vision: A Multi-Objective Reinforcement Learning Framework for Text-to-Visualization (2026.eacl-long)
Copied to clipboard
| Challenge: | Text2Vis systems generate functional code but resulting charts lack semantic alignment and clarity. |
| Approach: | They propose a framework that integrates post-execution feedback with textual accuracy, code validity, and visualization quality. |
| Outcome: | The proposed framework outperforms strong zero-shot and supervised baselines and shows robust generalization to out-of-domain datasets. |
Masked Thought: Simply Masking Partial Reasoning Steps Can Improve Mathematical Reasoning Learning of Language Models (2024.acl-long)
Copied to clipboard
Changyu Chen, Xiting Wang, Ting-En Lin, Ang Lv, Yuchuan Wu, Xin Gao, Ji-Rong Wen, Rui Yan, Yongbin Li
| Challenge: | Despite the advances in large language models, they still face difficulties with multi-step reasoning tasks. |
| Approach: | They propose a method that randomly masks certain tokens within the chain of thought to improve model accuracy by 5% over standard supervised fine-tuning. |
| Outcome: | The proposed method improves accuracy and accuracy by 5% over standard fine-tuning with a few codes modified. |
Select2Reason: Efficient Instruction-Tuning Data Selection for Long-CoT Reasoning (2026.findings-acl)
Copied to clipboard
| Challenge: | Large reasoning models exhibit human-like behaviors such as exploration, verification, reflection, and correction. |
| Approach: | They propose a supervised fine-tuning framework for long chain-of-thoughts reasoning . they leverage a difficulty-aware reward model to estimate the learning value of questions . |
| Outcome: | The proposed framework performs fine-tuning on large reasoning models on 10% of the data selected. |
SaySelf: Teaching LLMs to Express Confidence with Self-Reflective Rationales (2024.emnlp-main)
Copied to clipboard
| Challenge: | Existing approaches to elicit confidence from large language models are limited to binary or inaccurate group-level confidence estimates. |
| Approach: | They propose a training framework that teaches LLMs to express more fine-grained confidence estimates. |
| Outcome: | The proposed training framework reduces the confidence calibration error and maintains the performance of the model. |
PIC: Unlocking Long-Form Text Generation Capabilities of Large Language Models via Position ID Compression (2025.acl-long)
Copied to clipboard
| Challenge: | Long-context understanding is crucial for large language models (LLMs) however, the ability to “output-long” is underexplored. |
| Approach: | They propose a position ID compression approach to unlock the long-form text generation potential of large language models (LLMs). |
| Outcome: | The proposed approach can extend LLMs' generation length by 1.5 times without compromising generation quality. |
Selective Self-to-Supervised Fine-Tuning for Generalization in Large Language Models (2025.findings-naacl)
Copied to clipboard
| Challenge: | Large Language Models (LLMs) can be fine-tuned on task-specific data to improve performance on target tasks but can be overfitted resulting in a loss of generalization. |
| Approach: | They propose a method that uses the correct model responses from a training set to fine-tune the model using the correct response and the gold response for the remaining samples. |
| Outcome: | The proposed approach reduces model specialization during the fine-tuning stage while improving generalization. |
ILSIC: Corpora for Identifying Indian Legal Statutes from Queries by Laymen (2026.findings-eacl)
Copied to clipboard
| Challenge: | Existing studies have focused on the use of court judgments as input for legal Statute Identification (LSI) however, there is little research to explore the differences between court and laypeople data for LSI. |
| Approach: | They create a corpus of laypeople queries covering 500+ statutes from Indian law . they use court case judgements to compare between the two datasets . |
| Outcome: | The proposed corpus of laypeople queries covers 500+ statutes from Indian law . the results show that models trained on court judgements are ineffective . |
Learning from Relevant Subgoals in Successful Dialogs using Iterative Training for Task-oriented Dialog Systems (2024.findings-emnlp)
Copied to clipboard
| Challenge: | Task-oriented Dialog (ToD) systems have to solve multiple subgoals to accomplish user goals, whereas feedback is often obtained only at the end of the dialog. |
| Approach: | They propose an iterative training approach that uses subgoals to improve task-oriented dialog systems. |
| Outcome: | The proposed approach improves on a popular ToD benchmark by combining fine-tuning and preference learning steps. |
How to Train Long-Context Language Models (Effectively) (2025.acl-long)
Copied to clipboard
| Challenge: | a new study shows that language models can process extremely long contexts with minimal training. |
| Approach: | They use supervised fine-tuning and continued training to evaluate a language model's long-context capabilities. |
| Outcome: | The proposed model outperforms Llama-3.1-8B-Instruct on most long-context tasks . the model can process 512K tokens, one of the longest context windows of LMs . |
Persona-Consistent Dialogue Generation via Pseudo Preference Tuning (2025.coling-main)
Copied to clipboard
| Challenge: | Existing methods for improving persona consistency in dialogues require external resources. |
| Approach: | They propose a method for enhancing persona consistency in dialogue response generation using direct preference optimization using persona data. |
| Outcome: | The proposed method produces more consistent and natural responses than previous methods. |
The Reasoning Trap: How Enhancing LLM Reasoning Amplifies Tool Hallucination (2026.acl-long)
Copied to clipboard
| Challenge: | Recent studies suggest that strengthening reasoning often coincides with increased hallucination . however, no prior work has examined whether reasoning enhancement itself causes tool hallucinism . |
| Approach: | They propose a diagnostic benchmark measuring tool hallucination in two failure modes . they demonstrate a causal relationship between enhancing reasoning and tool hallubulation . |
| Outcome: | The proposed benchmark measures tool hallucination in two failure modes: no tool available, and (ii) only distractor tools available. |
NOVER: Incentive Training for Language Models via Verifier-Free Reinforcement Learning (2025.emnlp-main)
Copied to clipboard
| Challenge: | Recent advances in reinforcement learning, such as DeepSeek R1-Zero, highlight the effectiveness of incentive training, but these methods rely on external verifiers, which limits their applicability to domains like mathematics and coding, where such verifier is readily available. |
| Approach: | They propose a general reinforcement learning framework that requires only standard supervised fine-tuning data with no need for an external verifier. |
| Outcome: | The proposed framework outperforms the model of the same size distilled from large reasoning models such as DeepSeek R1 671B by 7.7%. |
NAT: Enhancing Agent Tuning with Negative Samples (2025.naacl-long)
Copied to clipboard
| Challenge: | Existing methods for fine-tuning and reinforcement learning use only positive examples, limiting their efficiency in low-resource scenarios. |
| Approach: | They propose a method that leverages both successful and failed trajectories for fine-tuning, maximizing the utility of limited resources. |
| Outcome: | The proposed method surpasses existing methods, including SFT, DPO, and PPO, across various tasks. |
ChiMed-GPT: A Chinese Medical Large Language Model with Full Training Regime and Better Alignment to Human Preferences (2024.acl-long)
Copied to clipboard
| Challenge: | Current large language models (LLMs) are ineffective in learning domain knowledge and aligning with human preference. |
| Approach: | They propose a benchmark LLM for Chinese medical domain that uses pre-training, supervised fine-tuning and RLHF to train LLMs. |
| Outcome: | The proposed LLM performs better than existing LLMs in the Chinese medical domain. |
Hidden Ghost Hand: Unveiling Backdoor Vulnerabilities in MLLM-Powered Mobile GUI Agents (2025.findings-emnlp)
Copied to clipboard
Pengzhou Cheng, Haowen Hu, Zheng Wu, Zongru Wu, Tianjie Ju, Daizong Ding, Zhuosheng Zhang, Gongshen Liu
| Challenge: | MLLM-powered GUI agents expose multiple interaction-level triggers, causing backdoor attacks . backdoor injection maximizes feature difference across sample classes, improving flexibility . |
| Approach: | They propose a framework for red-teaming backdoor attacks using MLLMs . they construct composite triggers by combining goal and interaction levels . |
| Outcome: | The proposed framework is effective and stealthy for red-teaming backdoor attacks. |
Large Language Models with Reinforcement Learning from Human Feedback Approach for Enhancing Explainable Sexism Detection (2025.coling-main)
Copied to clipboard
| Challenge: | Recent advances in natural language processing have significantly improved text comprehension. |
| Approach: | They propose a Reinforcement Learning from Human Feedback (RLHF) based fine-tuning framework for sexism detection that leverages contextual learning to understand and apply instructions to new scenarios without additional training. |
| Outcome: | The proposed framework outperforms existing models on three EDOS tasks and scores 0.8681 on binary sexism detection, 0.6829 on category classification of sexists and 0.4722 on task C. |
Warmup Generations: A Task-Agnostic Approach for Guiding Sequence-to-Sequence Learning with Unsupervised Initial State Generation (2025.acl-long)
Copied to clipboard
| Challenge: | Existing supervised fine-tuning (SFT) methods focus on directly generating the target output without leveraging the benefits of intermediate steps or initial guidance. |
| Approach: | They propose a task-agnostic framework that enables models to generate intermediate "warmup" sequences that are iteratively refined to maximize their contribution to the final output. |
| Outcome: | The proposed framework outperforms traditional supervised fine-tuning methods on translation, summarization, and multi-choice question answering tasks. |
Does Fine-Tuning LLMs on New Knowledge Encourage Hallucinations? (2024.emnlp-main)
Copied to clipboard
| Challenge: | Pre-training Large Language Models (LLMs) on textual corpora embeds substantial factual knowledge in their parameters, which is essential for excelling in various downstream applications. |
| Approach: | They propose to use supervised fine-tuning to align large language models to new factual information that is not acquired through pre-training. |
| Outcome: | The proposed model is trained to generate facts that are not grounded in pre-existing knowledge, but hallucinates when examples with new knowledge are learned. |
Unlocking Implicit Experience: Synthesizing Tool-Use Trajectories from Text (2026.acl-long)
Copied to clipboard
| Challenge: | Large language models (LLMs) can be used to effectively utilize tools in multi-turn interactions, but acquiring diverse and realistic multi-step tool-use data remains a challenge. |
| Approach: | They propose a text-based data synthesis pipeline that generates multi-turn tool-use trajectories from text corpora using relevance filtering, workflow tool extraction, trajectory grounding, and complexity refinement. |
| Outcome: | The proposed model achieves 14.9% improvement on the BFCL V3 Multi-turn benchmark while significantly reducing inference latency and costs. |
PA-RAG: RAG Alignment via Multi-Perspective Preference Optimization (2025.naacl-long)
Copied to clipboard
| Challenge: | Existing approaches to optimize RAG generators fail to align with RAG requirements thoroughly. |
| Approach: | They propose a method for optimizing the RAG generator from multiple preference perspectives to align with RAG requirements comprehensively. |
| Outcome: | The proposed method improves the performance of RAG generators by incorporating retrieved documents into the prompt. |
Supervised Optimism Correction: Be Confident When LLMs Are Sure (2025.findings-acl)
Copied to clipboard
| Challenge: | Recent advances in Large Language Models (LLMs) have demonstrated remarkable success across diverse tasks such as instruction following, code generation, and medical diagnosis. |
| Approach: | They propose a supervised fine-tuning-based auxiliary loss for Q-value estimations during supervised refinement. |
| Outcome: | The proposed method outperforms beam search on GSM8K, MATH, and GAOKAO on reasoning benchmarks. |
MuSe: Multi-Stage Graph Reasoning via Vision-Language Models (2026.acl-long)
Copied to clipboard
| Challenge: | Graph Neural Networks (GNNs) and graph transformers are inadequate for tasks with limited generalization. |
| Approach: | They propose a multi-stage graph reasoning framework based on vision-language models that incrementally samples and visualizes task-relevant subgraphs. |
| Outcome: | The proposed framework outperforms existing benchmarks in Graph-related tasks. |
On Learning to Summarize with Large Language Models as References (2024.naacl-long)
Copied to clipboard
Yixin Liu, Kejian Shi, Katherine He, Longtian Ye, Alexander Fabbri, Pengfei Liu, Dragomir Radev, Arman Cohan
| Challenge: | Recent studies have found that summaries generated by large language models (LLMs) are favored by human annotators when compared to reference summary from widely used summarization datasets. |
| Approach: | They propose to use large language models (LLMs) as reference learning settings for smaller text summarization models to investigate whether their performance can be substantially improved. |
| Outcome: | The proposed model outperforms standard supervised fine-tuning and human evaluations while retaining human-level performance. |
Weak-to-Strong Reasoning (2024.findings-emnlp)
Copied to clipboard
| Challenge: | Existing approaches to supervise large language models (LLMs) exceed human capabilities, but the effectiveness of this approach is still unexplored. |
| Approach: | They propose a weak-to-strong reasoning framework that enables strong models to refine training data . they use supervised fine-tuning and preference optimization to optimize weak models . |
| Outcome: | The proposed framework improves the reasoning capabilities of a language model using three weak models. |
MetaKP: On-Demand Keyphrase Generation (2024.findings-emnlp)
Copied to clipboard
| Challenge: | Existing keyphrase prediction methods only output a single set of keyphrases per document . however, existing methods fail to cater to diverse needs of users and downstream applications . |
| Approach: | They propose a method that requires keyphrases that conform to specific high-level goals or intents to generate on-demand keyphrase generation. |
| Outcome: | The proposed method surpasses the performance of a fully fine-tuned BART-base model in 0.548 SemF1 . it can be used in epidemic event detection from social media. |
LIONs: An Empirically Optimized Approach to Align Language Models (2024.emnlp-main)
Copied to clipboard
| Challenge: | Recent studies have focused on aligning large language models with pre-trained datasets. |
| Approach: | They conduct a rigorous analysis of a three-stage training pipeline using sequence packing, loss masking and increasing the preference dataset size in DPO to improve the performance of language models. |
| Outcome: | The proposed models outperform the official instruct models tuned with closed-source data and algorithms. |
Unlocking the Potential of Model Merging for Low-Resource Languages (2024.findings-emnlp)
Copied to clipboard
| Challenge: | Adapting large language models (LLMs) to new languages requires continual pre-training followed by supervised fine-tuning. |
| Approach: | They propose a model merging solution that integrates LLMs with distinct capabilities into a single model without additional training. |
| Outcome: | The proposed model merging outperforms CT-then-SFT in low-resource languages with scarce data. |
Aligning LLMs with Individual Preferences via Interaction (2025.coling-main)
Copied to clipboard
| Challenge: | Existing studies on LLMs alignment focus on generalizing their behavior to generalized values such as helpfulness, harmlessness, and honesty. |
| Approach: | They train large language models to "interact to align" to implicitly infer user preferences . they use a multi-turn preference dataset to generate a personalized alignment . |
| Outcome: | The proposed method enables dynamic, personalized alignment via interaction with a multi-turn preference dataset. |
When Long Helps Short: How Context Length in Supervised Fine-tuning Affects Behavior of Large Language Models (2025.emnlp-main)
Copied to clipboard
| Challenge: | Large language models (LLMs) have achieved impressive performance across NLP tasks. |
| Approach: | They propose to use long-context SFT to improve short-contemporary performance . they also decouple and analyze two key components, Multi-Head Attention and Feed-Forward Network . |
| Outcome: | The proposed model improves short-context performance, contrary to pretraining. |
How Likely Do LLMs with CoT Mimic Human Reasoning? (2025.coling-main)
Copied to clipboard
| Challenge: | Using chain-of-thought to elicit reasoning capabilities is not always effective and accurate. |
| Approach: | They compare the reasoning process of LLMs with humans to understand the causal chain . they find that LLM deviates from the ideal causal chain, resulting in spurious correlations . |
| Outcome: | The proposed method does not improve performance or accurately represent reasoning processes in LLMs. |
VulLibGen: Generating Names of Vulnerability-Affected Packages via a Large Language Model (2024.acl-long)
Copied to clipboard
Tianyu Chen, Lin Li, ZhuLiuchuan ZhuLiuchuan, Zongyang Li, Xueqing Liu, Guangtai Liang, Qianxiang Wang, Tao Xie
| Challenge: | Existing work on affected package identification is limited by large language models . a recent study shows that 84% third-party packages contain security vulnerabilities . |
| Approach: | They propose a method to use LLM to generate the affected package . they propose supervised fine-tuning, retrieval augmented generation and a local search algorithm . |
| Outcome: | The proposed method has an average precision of 0.806 for identifying vulnerable packages in four most popular ecosystems in GitHub Advisory. |
MOA: Multi-Objective Alignment for Role-Playing Agents (2026.acl-long)
Copied to clipboard
| Challenge: | Prior work on role-playing agents relies on supervised fine-tuning or reinforcement learning with scalarized rewards, but these approaches do not address the coordination of multiple reward dimensions during optimization. |
| Approach: | They propose a reinforcement-learning framework that enables multi-dimensional, fine-grained rubric optimization for general RPAs. |
| Outcome: | Experiments on PersonaGym and RoleMRC show that MOA improves multi-dimensional role-playing performance over supervised and standard RL baselines. |
Improving NMT Models by Retrofitting Quality Estimators into Trainable Energy Loss (2025.coling-main)
Copied to clipboard
| Challenge: | Reinforcement learning has shown great promise in aligning language models with human preferences in a variety of text generation tasks, including machine translation. |
| Approach: | They propose a method that employs quality estimators as trainable loss networks to backpropagate to the NMT model. |
| Outcome: | The proposed method outperforms strong baselines and proximal policy optimizations on English-to-Mongolian translation. |
Towards Implicit Bias Detection and Mitigation in Multi-Agent LLM Interactions (2024.findings-emnlp)
Copied to clipboard
| Challenge: | a recent study shows that large language models are susceptible to societal biases due to their exposure to human-generated data. |
| Approach: | They propose two strategies to mitigate implicit gender biases in large language models . they create scenarios where implicit gender is present and develop a metric to assess the presence of biase . |
| Outcome: | The proposed methods mitigate implicit biases with self-reflection and fine-tuning. |
LLM-SLM Collaborative Framework of Idiomatic Expression Generation (2026.acl-long)
Copied to clipboard
| Challenge: | Existing methods for idiomatic expression generation lack parallel data and manual annotations. |
| Approach: | They propose an iterative LLM-SLM collaborative framework that replaces human supervision for idiomatic expression data generation. |
| Outcome: | The proposed framework outperforms DeepSeek-R1 in Chinese Idiom Polishing with a 25.2% improvement in accuracy. |
Time-RA: Towards Time Series Reasoning for Anomaly Diagnosis with LLM Feedback (2026.findings-acl)
Copied to clipboard
Yiyuan Yang, Zichuan Liu, Lei Song, Kai Ying, Stephen Wang, Joshua Thomas Bamford, Svitlana Vyetrenko, Jiang Bian, Qingsong Wen
| Challenge: | Time series anomaly detection (TSAD) has traditionally focused on binary classification and lacks the fine-grained categorization and explanatory reasoning required for transparent decision-making. |
| Approach: | They propose a time-series reasoning task that reformulates TSAD from discriminative to reasoning-intensive paradigm. |
| Outcome: | The proposed task reformulates TSAD from discriminative to reasoning-intensive paradigm. |
Bootstrapping LLM-based Task-Oriented Dialogue Agents via Self-Talk (2024.findings-acl)
Copied to clipboard
| Challenge: | Large language models (LLMs) are powerful dialogue agents, but specializing them towards fulfilling a specific function can be prohibitive in terms of feasibility, time, and resources. |
| Approach: | They propose a method for training large language models by enabling "self-talk" they propose supervised fine-tuning of LLMs to improve quality of dialogues . |
| Outcome: | The proposed method generates training data via "self-talk" of LLMs that can be refined and utilized for supervised fine-tuning. |
Can Textual Unlearning Solve Cross-Modality Safety Alignment? (2024.findings-emnlp)
Copied to clipboard
Trishna Chakraborty, Erfan Shayegani, Zikui Cai, Nael Abu-Ghazaleh, M. Salman Asif, Yue Dong, Amit Roy-Chowdhury, Chengyu Song
| Challenge: | integrating new modalities into large language models creates new attack surface . existing safety training techniques like SFT and RLHF are not feasible in multi-modal settings . |
| Approach: | They explore whether unlearning in the textual domain can be effective for cross-modality safety alignment. |
| Outcome: | The proposed approach reduces the Attack Success Rate (ASR) to less than 8% and preserves the utility. |
A Chain-of-Task Framework for Instruction Tuning of LLMs Based on Chinese Grammatical Error Correction (2025.coling-main)
Copied to clipboard
| Challenge: | Existing approaches to address Grammatical Error Correction (GEC) tasks are based on large scale labeled data, which leads to extremely high data annotation costs. |
| Approach: | They propose a Chain-of-Task framework to reduce over-correction in large language models . they propose supervised fine-tuning strategy and an algorithm for automatic dataset annotation . |
| Outcome: | The proposed framework achieves state-of-the-art on both FCGEC (in-domain) and NaCGEC (out-of domain) test sets. |
Stimulate the Critical Thinking of LLMs via Debiasing Discussion (2025.emnlp-main)
Copied to clipboard
| Challenge: | Existing studies show that large language models (LLMs) are often prone to stance homogeneity and human preference biases when faced with conflicting perspectives. |
| Approach: | They propose a novel two-stage training framework to address stance homogeneity bias and human preference bias by generating multi-model discussion datasets and optimizing reinforcement learning from human feedback to align with discussion correctness. |
| Outcome: | The proposed framework reduces stance homogeneity bias and human preference bias and improves generalization capabilities on non-discussion scenarios and out-of-domain datasets. |
On the Impact of Fine-Tuning on Chain-of-Thought Reasoning (2025.naacl-long)
Copied to clipboard
| Challenge: | Large language models have emerged as powerful tools for general intelligence, showcasing advanced natural language processing capabilities. |
| Approach: | They propose to use supervised fine-tuning and Quantized Low-Rank Adapters to improve LLMs' task-specific performance to address privacy and safety risks. |
| Outcome: | The proposed model improves the accuracy of the chain-of-thought reasonings across four datasets and demonstrates that the faithfulness of CoT reasoning decreases. |
ACECODER: Acing Coder RL via Automated Test-Case Synthesis (2025.acl-long)
Copied to clipboard
| Challenge: | Recent coder models have been driven by supervised fine-tuning, but the potential of reinforcement learning remains unexplored due to the lack of reliable reward data/model in the code domain. |
| Approach: | They propose a pipeline that generates extensive test-case pairs from existing code data and constructs preference pairs based on pass rates over sampled programs. |
| Outcome: | The proposed pipeline generates extensive (question, test-cases) pairs from existing code data and trains them with Bradley-Terry loss. |
PromptOptMe: Error-Aware Prompt Compression for LLM-based MT Evaluation Metrics (2025.naacl-long)
Copied to clipboard
| Challenge: | Recent efforts to improve the quality of machine-generated natural language content have been limited due to the large token usage required by complex evaluation prompts. |
| Approach: | They propose a prompt optimization approach that uses a smaller, fine-tuned language model to compress input data for evaluation prompt, thus reducing token usage and computational cost when using larger LLMs for downstream evaluation. |
| Outcome: | The proposed approach reduces token usage and costs by 2.37 compared with larger LLMs for downstream evaluation. |
Uncertainty Aware Learning for Language Model Alignment (2024.acl-long)
Copied to clipboard
| Challenge: | Existing alignment strategies that focus on diverse and high-quality data often overlook the intrinsic uncertainty of tasks, learning all data samples equally. |
| Approach: | They propose to introduce the sample uncertainty into the alignment of different task scenarios by a simple fashion by setting the label smoothing value of training according to the uncertainty of individual samples. |
| Outcome: | The proposed model outperforms standard supervised fine-tuning on high-entropy tasks and complex low-entropic tasks. |
DREAM: Disentangling Risks to Enhance Safety Alignment in Multimodal Large Language Models (2025.naacl-long)
Copied to clipboard
Jianyu Liu, Hangyu Guo, Ranjie Duan, Xingyuan Bu, Yancheng He, Shilong Li, Hui Huang, Jiaheng Liu, Yucheng Wang, Chenchen Jing, Xingwei Qu, Xiao Zhang, Pei Wang, Yanan Wu, Jihao Gu, Yangguang Li, Jianke Zhu
| Challenge: | Multimodal Large Language Models (MLLMs) pose unique safety challenges due to their integration of visual and textual data. |
| Approach: | They propose a method to disentangle risks through step-by-step reasoning within multimodal inputs. |
| Outcome: | The proposed approach improves safety alignment in MLLMs by fine-tuning and iterative Reinforcement Learning from AI feedback. |
Balancing Classification and Calibration Performance in Decision-Making LLMs via Calibration Aware Reinforcement Learning (2026.findings-acl)
Copied to clipboard
| Challenge: | Large language models (LLMs) are increasingly deployed in decision-making tasks where accuracy and reliable confidence estimates are essential. |
| Approach: | They propose a calibration-aware reinforcement learning formulation that directly adjusts decision-token probabilities. |
| Outcome: | The proposed model preserves RLVR’s accuracy level while mitigating overconfidence, reducing ECE scores up to 9 points. |
Learning While Staying Curious: Entropy-Preserving Supervised Fine-Tuning via Adaptive Self-Distillation for Large Reasoning Models (2026.acl-long)
Copied to clipboard
Hao Wang, Hao Gu, Hongming Piao, Kaixiong Gong, Yuxiao Ye, Xiangyu Yue, Sirui Han, Yike Guo, Dapeng Wu
| Challenge: | Recent advances establish "SFT-then-RL" as the defacto paradigm for enhancing large reasoning mod- els on automatically verifiable tasks. |
| Approach: | They propose an entropy-preserving SFT method to enhance exploration capabilities through intrinsic curiosity. |
| Outcome: | The proposed method outperforms the vanilla method on reasoning tasks by 2.5 points . it also outperformed the vanilla SFT by 2.9 points on out-of-distribution tasks . |
Correct, Concise and Complete: Multi-stage Training For Adaptive Reasoning (2026.findings-acl)
Copied to clipboard
| Challenge: | Large language models (LLMs) increase test-time computation, often in the form of chain-of-thought (CoT) however, reasoning traces can become unnecessarily long, increasing computation costs without improving accuracy and sometimes even degrading performance. |
| Approach: | They propose a multi-stage efficient reasoning method that combines supervised fine-tuning with reinforcement learning using an adaptive length penalty. |
| Outcome: | The proposed method reduces response length by an average of 28% for 8B models and 40% for 32B models while incurring only minor performance drops of 1.6 and 2.5 points, respectively. |
ORPO: Monolithic Preference Optimization without Reference Model (2024.emnlp-main)
Copied to clipboard
| Challenge: | Pre-trained language models with vast training corpora have shown remarkable abilities in diverse natural language processing tasks. |
| Approach: | They propose a model-free monolithic odds ratio preference optimization algorithm, ORPO, to improve preference alignment. |
| Outcome: | The proposed algorithm outperforms state-of-the-art language models with more than 7B and 13B parameters on the ultrafeedback alone. |
HFT: Half Fine-Tuning for Large Language Models (2025.acl-long)
Copied to clipboard
| Challenge: | Large language models (LLMs) with one or more fine-tuning phases can unlock various capabilities, but can be catastrophic forgetting during sequential training. |
| Approach: | They propose a method to regularly reset partial parameters to mitigate forgetting issues by using half fine-tuning instead of full fine-uning. |
| Outcome: | The proposed approach reduces the risk of catastrophic forgetting during training and the parametric knowledge lost during training may be overwhelmed by incoming training data. |
Towards Adapting Open-Source Large Language Models for Expert-Level Clinical Note Generation (2025.findings-acl)
Copied to clipboard
Hanyin Wang, Chufan Gao, Bolun Liu, Qiping Xu, Guleid Hussein, Mohamad El Labban, Kingsley Iheasirim, Hariprasad Reddy Korsapati, Chuck Outcalt, Jimeng Sun
| Challenge: | Proprietary Large Language Models (LLMs) have demonstrated promising capabilities in clinical text summarization tasks. |
| Approach: | They propose a domain- and task-specific adaptation process for an open-source LLaMA-2 model . LLama-2 can generate high-quality clinical notes from outpatient patient-doctor dialogues . |
| Outcome: | The proposed model can generate clinical notes comparable to those authored by physicians. |
Lost in Literalism: How Supervised Training Shapes Translationese in LLMs (2025.acl-long)
Copied to clipboard
| Challenge: | Large language models exhibit translationese errors and generate unexpected unnatural translations . Neural machine translation (NMT) has become the dominant method in machine translation research . |
| Approach: | They evaluate the prevalence of translationese in LLM-generated translations and investigate its roots during supervised fine-tuning. |
| Outcome: | The proposed methods reduce translationese while improving translation naturalness . the proposed methods are validated by human evaluations and automatic metrics . |
ExecVerify: White-Box RL with Verifiable Stepwise Rewards for Code Execution Reasoning (2026.acl-long)
Copied to clipboard
| Challenge: | Existing methods for code execution reasoning are limited by the difficulty of the training data. |
| Approach: | They propose a model that uses reinforcement learning to reward correct answers from execution traces. |
| Outcome: | The proposed model improves pass@1 by up to 5.9% on code generation tasks over strong baselines. |
TRANS-ZERO: Self-Play Incentivizes Large Language Models for Multilingual Translation Without Parallel Data (2025.findings-acl)
Copied to clipboard
| Challenge: | Large Language Models (LLMs) have reshaped machine translation, but multilingual MT still relies heavily on parallel data for supervised fine-tuning. |
| Approach: | They propose a framework that leverages only monolingual data and the intrinsic multilingual knowledge of Large Language Models (LLMs). |
| Outcome: | The proposed framework matches models trained on large-scale parallel data and excels in non-English translation directions. |
Curr-ReFT: Overcoming Training Bottlenecks in Small-scale Vision-Language Models via Curriculum Reinforcement Finetuning (2025.findings-emnlp)
Copied to clipboard
| Challenge: | State-of-the-art vision-language models require massive scaling that limits practical deployment. |
| Approach: | They propose to use supervised fine-tuning to train small-scale vision-language models but face out-of-domain collapse when trained with traditional supervised learning (SFT). |
| Outcome: | Experiments show that curr-reFT achieves state-of-the-art performance across visual tasks in both in- and out-of domain settings and benchmarks. |
CLaw: Benchmarking Chinese Legal Knowledge in Large Language Models - A Fine-grained Corpus and Reasoning Analysis (2025.findings-emnlp)
Copied to clipboard
| Challenge: | a new benchmark is designed to evaluate LLMs on Chinese legal knowledge and its application in reasoning . general pre-training that ingests legal texts without specialized focus compromises reliability of LLM responses . achieving trustworthy legal reasoning in LLM requires a robust synergy of accurate knowledge retrieval and strong general reasoning capabilities. |
| Approach: | They propose a benchmark specifically engineered to evaluate LLMs on Chinese legal knowledge and its application in reasoning. |
| Outcome: | The proposed benchmark evaluates LLMs on Chinese legal knowledge and its application in reasoning. |
Reflective RAG: Self-Evaluation Driven Strategy Optimization in Agentic Retrieval-Augmented Generation (2026.findings-acl)
Copied to clipboard
| Challenge: | Recent agentic RAG systems lack the capacity to evaluate the utility of retrieved information, leading to brittle reasoning and suboptimal decision-making. |
| Approach: | They propose a framework that integrates self-evaluation to dynamically optimize retrieval and generation strategy. |
| Outcome: | The proposed framework outperforms strong agentic baselines on five knowledge-intensive QA benchmarks and improves training stability and generalization to multi-hop reasoning tasks. |
Enhancing Nursing and Elderly Care with Large Language Models: An AI-Driven Framework (2025.coling-main)
Copied to clipboard
| Challenge: | Experimental results demonstrate significant improvements, paving the way for AI-driven solutions to meet the growing demands of healthcare in aging populations. |
| Approach: | They introduce a Chinese nursing dataset and implement incremental pre-training and supervised fine-tuning techniques to enhance LLM performance in specialized tasks. |
| Outcome: | The proposed model performs better in real-time patient monitoring and interaction tasks than previous models. |
CodeAttack: Revealing Safety Generalization Challenges of Large Language Models via Code Completion (2024.findings-acl)
Copied to clipboard
| Challenge: | Large Language Models exhibit remarkable generative capabilities but can be misused for harmful purposes. |
| Approach: | They propose a framework that transforms natural language inputs into code inputs. |
| Outcome: | The proposed framework bypasses the safety guardrails of all models more than 80% of the time. |
Large Language Models Can Not Perform Well in Understanding and Manipulating Natural Language at Both Character and Word Levels? (2024.findings-emnlp)
Copied to clipboard
| Challenge: | Large language models (LLMs) still exhibit significant deficiencies in basic language understanding and manipulation. |
| Approach: | They propose a bilingual benchmark to assess the performance of Large language models . they use a set of 15 simple text editing tasks to examine their capabilities . |
| Outcome: | The proposed benchmark aims to assess the performance of Large language models in basic language tasks. |
Few-Shot Open-Set Classification via Reasoning-Aware Decomposition (2025.emnlp-main)
Copied to clipboard
| Challenge: | Large language models excel at few-shot learning, but their ability to reject out-of-distribution examples remains under-explored. |
| Approach: | They introduce a novel amortised Generative Flow Network framework that uses latent trajectories to approximate the Bayesian posterior. |
| Outcome: | The proposed framework can generalise with as few as 4 examples per class, enabling Llama 3.2 3B to achieve up to 80% of the performance of Llma 3.3 70B in complex datasets. |
A Two-Stage Adaptation of Large Language Models for Text Ranking (2024.findings-acl)
Copied to clipboard
| Challenge: | Recent advances in pre-trained language models (PLMs) have significantly improved ranking performance in text ranking tasks. |
| Approach: | They propose a two-stage progressive paradigm to better adapt LLMs to text ranking by conducting continual pre-training on a large weakly-supervised corpus and performing SFT on high-quality data. |
| Outcome: | The proposed approach outperforms previous methods on in- and out-domain scenarios. |
Teaching Your Models to Understand Code via Focal Preference Alignment (2025.emnlp-main)
Copied to clipboard
Jie Wu, Haoling Li, Xin Zhang, Xiao Liu, Yangyu Huang, Jianwen Luo, Yizhen Zhang, Zuchao Li, Ruihang Chu, Yujiu Yang, Scarlett Li
| Challenge: | Existing methods for supervised fine-tuning focus on unit test feedback to construct preference pairs. |
| Approach: | They propose a preference alignment framework that mimics human iterative debugging to refine Code LLMs. |
| Outcome: | Experiments show that Preference Learning improves on BigCodeBench and BigCodeBind tasks. |
Pushing the Limits of LLM Tool Calling via Experiential Knowledge Integration and Activation (2026.findings-acl)
Copied to clipboard
| Challenge: | Existing approaches treat tool use as a problem of prompt design, API documents specification, or supervised or unsupervised alignment. |
| Approach: | They propose a knowledge-augmented tool execution framework that integrates experiential knowledge with reasoning-width-expanded inference and knowledge-aware training. |
| Outcome: | The proposed framework improves on BFCL-V3 and AppWorld on three model scales. |
LPO: Towards Accurate GUI Agent Interaction via Location Preference Optimization (2026.findings-acl)
Copied to clipboard
Jiaqi Tang, Yu Xia, Yi-Feng Wu, Yuwei Hu, Chen Yuhui, Qing-Guo Chen, Xiaogang Xu, Xiangyu Wu, Hao LU, Yanqing Ma, Shiyin Lu, Qifeng Chen
| Challenge: | Existing strategies for spatial localization are limited due to their limited capacity to perceive positional data. |
| Approach: | They propose a location-based approach that leverages locational data to optimize interaction preferences. |
| Outcome: | The proposed approach achieves SOTA results across offline benchmarks and real-world evaluations. |
Prefix Text as a Yarn: Eliciting Non-English Alignment in Foundation Language Model (2024.findings-acl)
Copied to clipboard
| Challenge: | supervised fine-tuning (SFT) has been a straightforward approach for tailoring the output of foundation large language models (LLMs) to specific preferences. |
| Approach: | They propose a training-free alignment method that uses minimal prior tokens to bridge the foundation LLM and the SFT LLM. |
| Outcome: | The proposed method achieves comparable performance without training on machine translation and part-of-speech tagging across seven languages. |
TRIDENT: Enhancing Large Language Model Safety with Tri-Dimensional Diversified Red-Teaming Data Synthesis (2025.acl-long)
Copied to clipboard
| Challenge: | Large Language Models (LLMs) excel in natural language processing tasks but are vulnerable to harmful content and being exploited for malicious purposes. |
| Approach: | They propose a framework to measure the risk coverage of alignment datasets across three dimensions: Lexical Diversity, Malicious Intent, and Jailbreak Tactics. |
| Outcome: | The proposed framework measures risk coverage across Lexical Diversity, Malicious Intent, and Jailbreak Tactics. |
Cross-Lingual Optimization for Language Transfer in Large Language Models (2025.acl-long)
Copied to clipboard
| Challenge: | Adapting large language models to other languages often suffers from an overemphasis on English performance. |
| Approach: | They propose a cross-lingual optimization technique that efficiently transfers an English-centric LLM to a target language while preserving its English capabilities. |
| Outcome: | The proposed model outperforms SFT in acquiring target language proficiency and maintaining English performance in low-resource languages. |
LoRA-PAR: A Flexible Dual-System LoRA Partitioning Approach to Efficient LLM Fine-Tuning (2025.findings-emnlp)
Copied to clipboard
| Challenge: | Large-scale generative models like DeepSeek-R1 and OpenAI-O1 benefit substantially from chain-of-thought reasoning, yet pushing their performance typically requires vast data, large model sizes, and full-parameter fine-tuning. |
| Approach: | They propose a dual-system LoRA framework that partitions data and parameters by System 1 or System 2 demands and adopts a two-stage fine-tuning strategy to enhance knowledge and intuition. |
| Outcome: | The proposed framework partitions data and parameters by System 1 or System 2 demands, using fewer yet more focused parameters for each task. |
Exploring Multimodal Challenges in Toxic Chinese Detection: Taxonomy, Benchmark, and Findings (2025.findings-acl)
Copied to clipboard
Shujian Yang, Shiyao Cui, Chuanrui Hu, Haicheng Wang, Tianwei Zhang, Minlie Huang, Jialiang Lu, Han Qiu
| Challenge: | Recent studies show that character substitutions in toxic Chinese text can confuse state-of-the-art LLMs. |
| Approach: | They propose a taxonomy of 3 perturbation strategies and 8 specific approaches in Chinese text to assess if they can detect perturbed Chinese toxic contents. |
| Outcome: | The proposed model can detect perturbed Chinese text with 8 different approaches . the proposed model is compared with 9 other LLMs from the US and China . |
On Collaborating Small and Large Models For Few-shot Intent Detection (2025.findings-emnlp)
Copied to clipboard
| Challenge: | Existing methods for few-shot intent detection face high inference cost and label interference. |
| Approach: | They propose a framework that integrates a small prediction model with a large language model for FSID. |
| Outcome: | The proposed framework outperforms existing methods on three benchmark datasets. |
LlmLink: Dual LLMs for Dynamic Entity Linking on Long Narratives with Collaborative Memorisation and Prompt Optimisation (2025.coling-main)
Copied to clipboard
| Challenge: | Existing methods focus on supervised fine-tuning or limited to one-off prediction, which poses a challenge where the context is long. |
| Approach: | They propose a dynamic approach to CoREFerence resolution in chunked long narratives by deploying dual Large Language Models. |
| Outcome: | The proposed model achieves performance gains over existing models and fine-tuning approaches on long narrative datasets, significantly reducing the resources required for inference and training. |
SelfAug: Mitigating Catastrophic Forgetting in Retrieval-Augmented Generation via Distribution Self-Alignment (2025.findings-emnlp)
Copied to clipboard
Yuqing Huang, Rongyang Zhang, Qimeng Wang, Chengqiang Lu, Yan Gao, null Yiwu, Yao Hu, Xuyang Zhi, Guiquan Liu, Xin Li, Hao Wang, Enhong Chen
| Challenge: | Existing solutions for supervised fine-tuning often lead to catastrophic forgetting, where models lose their previously acquired knowledge and general capabilities. |
| Approach: | They propose a self-distribution alignment method that aligns input sequence logits to preserve the model’s semantic distribution, thereby mitigating catastrophic forgetting and improving downstream performance. |
| Outcome: | The proposed method achieves a superior balance between downstream learning and general capability retention. |
Enhancing Generative Retrieval with Reinforcement Learning from Relevance Feedback (2023.emnlp-main)
Copied to clipboard
| Challenge: | End-to-end generative retrieval models produce document identifiers in response to a query . however, this approach has two challenges: an overemphasis on top-1 results at the expense of overall ranking quality. |
| Approach: | They propose a generative retrieval model with reinforcement learning from relevance feedback to align token-level docid generation with document-level relevance estimation. |
| Outcome: | The proposed model aligns token-level docid generation with document-level relevance estimation. |
A Systematic Analysis of Large Language Models as Soft Reasoners: The Case of Syllogistic Inferences (2024.emnlp-main)
Copied to clipboard
| Challenge: | syllogistic reasoning is a deductive reasoning skill that is crucial in everyday problem-solving and decision-making experiences. |
| Approach: | They propose to study the reasoning abilities of Large Language Models (LLMs) they propose to use supervised fine-tuning and chain-of-thought reasoning to investigate their results. |
| Outcome: | The proposed models exhibit reasoning biases, avoid answering that no conclusion follows, align with human difficulties, and struggle with multi-step reasoning. |
Improving Low-Resource Keyphrase Generation through Unsupervised Title Phrase Generation (2024.lrec-main)
Copied to clipboard
| Challenge: | Existing methods for unsupervised keyphrase generation use phrases extracted from document title instead of phrase bank. |
| Approach: | They propose a method for generating pseudo labels from a document title . they use phrases mined from the document title to generate absent keyphrases . |
| Outcome: | The proposed method outperforms existing methods on human-annotated datasets even with fewer labeled data. |
Unveiling the Key Factors for Distilling Chain-of-Thought Reasoning (2025.findings-acl)
Copied to clipboard
Xinghao Chen, Zhijing Sun, Guo Wenjin, Miaoran Zhang, Yanjun Chen, Yirong Sun, Hui Su, Yijie Pan, Dietrich Klakow, Wenjie Li, Xiaoyu Shen
| Challenge: | Large Language Models (LLMs) excel in reasoning tasks through Chain-of-Thought prompting. |
| Approach: | They examine the factors influencing CoT distillation including granularity, format and teacher model. |
| Outcome: | The proposed model is based on four teacher models and seven student models across seven mathematical and commonsense reasoning datasets. |
Decoding Uncertainty: The Impact of Decoding Strategies for Uncertainty Estimation in Large Language Models (2025.findings-emnlp)
Copied to clipboard
| Challenge: | Decoding strategies affect the probability distribution underlying the output of a language model and can therefore affect both generation quality and uncertainty. |
| Approach: | They investigate the impact of decoding strategies on uncertainty estimation in large language models . |
| Outcome: | The proposed methods improve the uncertainty estimation of large language models by reducing repetition. |
TempTool-R1: Tool-Augmented Reinforcement Learning for Temporal Knowledge Graph Question Answering (2026.findings-acl)
Copied to clipboard
| Challenge: | Existing approaches to temporal knowledge graph question answering struggle with multi-hop reasoning and implicit temporal constraints. |
| Approach: | They propose a temporal tool-based API capable of transforming implicit temporal cues into executable operations and supervised fine-tuning teaches the model to interweave chain-of-thought reasoning with think-then-tool usage. |
| Outcome: | The proposed framework outperforms existing methods on three challenging questions. |
Pedagogical Alignment of Large Language Models (2024.findings-emnlp)
Copied to clipboard
| Challenge: | Large Language Models (LLMs) are often used without pedagogical fine-tuning and provide immediate answers rather than guiding students through the problem-solving process. |
| Approach: | They propose a method for constructing large-scale preference datasets using synthetic data generation techniques that eliminates the need for manual annotation. |
| Outcome: | The proposed methods outperform standard supervised fine-tuning (SFT) and improve alignment accuracy by 13.1% and 8.7% respectively. |
MMBoundary: Advancing MLLM Knowledge Boundary Awareness through Reasoning Step Confidence Calibration (2025.acl-long)
Copied to clipboard
| Challenge: | Existing methods calibrate model confidence on entire response, which leads to incorrect answers with high confidence. |
| Approach: | They propose a framework that advances the knowledge boundary awareness of multimodal large language models through reasoning step confidence calibration. |
| Outcome: | Empirical results show that the proposed framework outperforms existing methods across domains and metrics. |
ConstraintLLM: A Neuro-Symbolic Framework for Industrial-Level Constraint Programming (2025.emnlp-main)
Copied to clipboard
| Challenge: | Constraint programming (CP) is a powerful paradigm for solving constraint optimization problems. |
| Approach: | They propose to use an open-source LLM to generate formal modeling for COPs. |
| Outcome: | The proposed model outperforms the baselines on the new IndusCP benchmark by 2x. |
TRN-R1-Zero: Text-rich Network Reasoning via LLMs with Reinforcement Learning Only (2026.acl-long)
Copied to clipboard
| Challenge: | Recent large language model-based approaches often overlook graph context or depend on distillation from larger models, limiting generalisation. |
| Approach: | They propose a framework for zero-shot reasoning on text-rich networks . they use a Neighbour-aware Group Relative Policy Optimisation objective . |
| Outcome: | The proposed framework optimises base LLMs using a Neighbour-aware group relative policy optimisation objective based on a novel margin gain metric for the informativeness of neighbouring signals . |
Chain of Strategy Optimization Makes Large Language Models Better Emotional Supporter (2025.findings-emnlp)
Copied to clipboard
Weixiang Zhao, Xingyu Sui, Xinyang Han, Yang Deng, Yulin Hu, Jiahe Guo, Libo Qin, Qianyun Du, Shijin Wang, Yanyan Zhao, Bing Qin, Ting Liu
| Challenge: | Existing supervised fine-tuning (SFT) fails to address these issues, as it trains models on single gold-standard responses without modeling nuanced strategy trade-offs. |
| Approach: | They propose a two-stage framework that optimizes strategy selection preferences at each dialogue turn. |
| Outcome: | The proposed framework improves strategy selection preferences at each dialogue turn. |
CADMate: Generating CAD Assembly Plan with Geometric Chain-of-Thought and Spatial Physical Rewards (2026.acl-long)
Copied to clipboard
| Challenge: | Computer-aided design (CAD) is crucial in prototyping complex 3D objects . designers manually define assembly sequences for individual CAD parts . |
| Approach: | They propose a framework for computer-aided design that predicts actions for CAD parts . they use a reference design image and disassembled parts to generate 6-DoF transformations . |
| Outcome: | The proposed framework outperforms existing MLLMs in the design of CAD assemblies. |
Context-attended Adversarial Reinforcement Learning for Robust Multi-step Retrieval Augmented Generation (2026.findings-acl)
Copied to clipboard
| Challenge: | Existing approaches to multi-step retrieval-augmented generation are susceptible to retrieval noise and fabricated documents in real-world scenarios. |
| Approach: | They propose a framework for multi-step retrieval-augmented generation that incorporates external knowledge into a retriever to generate responses from adversarial samples. |
| Outcome: | The proposed framework improves performance in multiple noisy scenarios and can be used to improve multi-step retrieval-augmented generation. |
ReviewRL: Towards Automated Scientific Review with RL (2025.emnlp-main)
Copied to clipboard
Sihang Zeng, Kai Tian, Kaiyan Zhang, Yuru Wang, Junqi Gao, Runze Liu, Sa Yang, Jingxuan Li, Xinwei Long, Jiaheng Ma, Biqing Qi, Bowen Zhou
| Challenge: | Existing automated review systems struggle with factual accuracy, rating consistency, and analytical depth. |
| Approach: | They propose a framework for generating comprehensive and factually grounded scientific paper reviews using supervised fine-tuning and reinforcement learning. |
| Outcome: | The proposed framework outperforms existing methods on ICLR 2025 papers. |
Can Machines Resonate with Humans? Evaluating the Emotional and Empathic Comprehension of LMs (2024.findings-emnlp)
Copied to clipboard
| Challenge: | Empathy plays a pivotal role in fostering prosocial behavior, often triggered by the sharing of personal experiences through narratives. |
| Approach: | They propose to use contrastive learning with masked LMs and supervised fine-tuning with large language models to improve empathy understanding in NLP models. |
| Outcome: | The proposed methods show that there is low agreement among annotators and that cultural differences are a factor in their interpretation of empathy. |
ChatSOP: An SOP-Guided MCTS Planning Framework for Controllable LLM Dialogue Agents (2025.acl-long)
Copied to clipboard
Zhigen Li, Jianxiang Peng, Yanmeng Wang, Yong Cao, Tianhao Shen, Minghui Zhang, Linxi Su, Shang Wu, Yihang Wu, YuQian Wang, Ye Wang, Wei Hu, Jianfeng Li, Shaojun Wang, Jing Xiao, Deyi Xiong
| Challenge: | Existing models that use Large Language Models (LLMs) show superior performance in various tasks, but lack of controllability leads to unfocused conversations or task failure. |
| Approach: | They propose a standard operating procedure (SOP) framework to regulate dialogue flow by integrating Chain of Thought reasoning and supervised fine-tuning for SOP prediction. |
| Outcome: | The proposed method achieves a 27.95% improvement in action accuracy compared to baseline models based on GPT-3.5 and also shows notable gains for open-source models. |
Language Models Can Easily Learn to Reason from Demonstrations (2025.findings-emnlp)
Copied to clipboard
Dacheng Li, Shiyi Cao, Tyler Griggs, Shu Liu, Xiangxi Mo, Eric Tang, Sumanth Hegde, Kourosh Hakhamaneshi, Shishir G Patil, Matei Zaharia, Joseph E. Gonzalez, Ion Stoica
| Challenge: | Large reasoning models (LRMs) tackle complex problems by following long chain-of-thoughts (Long CoT) however, the training techniques and data requirements to elicit Long CoT remain poorly understood. |
| Approach: | They propose to use data-efficient supervised fine-tuning and parameter-efficient low-rank adaptation to elicit Long CoT reasoning. |
| Outcome: | The proposed model can learn Long CoT reasoning through data-efficient supervised fine-tuning and parameter-efficient low-rank adaptation. |
Sparsity-Accelerated Training for Large Language Models (2024.findings-acl)
Copied to clipboard
| Challenge: | Large language models (LLMs) have demonstrated proficiency across various NLP tasks but often require additional training, such as continual pre-training and supervised fine-tuning. |
| Approach: | They propose to leverage sparsity in pre-trained LLMs to accelerate training by disregarding computations for unimportant neurons. |
| Outcome: | The proposed framework achieves comparable or superior performance to standard training while significantly accelerating the process. |
Reinforcement Learning with Semantic Rewards Enables Low-Resource Language Expansion without Alignment Tax (2026.findings-acl)
Copied to clipboard
Zeli Su, Ziyin Zhang, Zhou Liu, Xuexian Song, Zhankai Xu, Longfei Zheng, Xiaolu Zhang, Rong Fu, Guixian Xu, Wentao Zhang
| Challenge: | Extending large language models to low-resource languages often incurs an "alignment tax" token-level fine-tuning enforces token-level surface imitation on narrow and biased data distributions. |
| Approach: | They propose a semantic-space alignment paradigm powered by group-level semantic rewards instead of likelihood maximization. |
| Outcome: | The proposed model acquires low-resource capa- bilities while mitigating alignment tax on Tibetan–Chinese machine translation and Ti- betan headline generation. |
TTPA: Token-level Tool-use Preference Alignment Training Framework with Fine-grained Evaluation (2025.findings-emnlp)
Copied to clipboard
| Challenge: | Existing tool-learning methods often overlook fine-grained optimization of internal tool call details. |
| Approach: | They propose a training paradigm for constructing token-level tool-use preference datasets . reversed dataset construction is a method for creating high-quality, multi-turn tool-user datasets by reversing the generation flow. |
| Outcome: | a new training paradigm improves tool-using performance and generalizes results. |
Making Reasoning Matter: Measuring and Improving Faithfulness of Chain-of-Thought Reasoning (2024.findings-emnlp)
Copied to clipboard
| Challenge: | Large language models (LLMs) are shown to perform better when asked to reason step-by-step before generating a final answer. |
| Approach: | They propose a framework to tailor small-sized LMs to generate correct reasoning steps and robustly reason over these steps. |
| Outcome: | The proposed framework outperforms four competitive baselines and improves the robustness and generalization ability of the reasoning LM, yielding higher performance on out-of-distribution test sets. |
CodeTool: Enhancing Programmatic Tool Invocation of LLMs via Process Supervision (2025.acl-long)
Copied to clipboard
YifeiLu YifeiLu, Fanghua Ye, Jian Li, Qiang Gao, Cheng Liu, Haibo Luo, Nan Du, Xiaolong Li, Feiliang Ren
| Challenge: | Existing approaches to tool invocation are often unnecessarily long and require lengthy reasoning paths. |
| Approach: | They propose a framework for stepwise code generation that improves LLM tool invocation . they incorporate two distinct process rewards: the On-the-spot and the Latent Reward . |
| Outcome: | The proposed framework improves LLM tool invocation by leveraging the concise nature of code. |
Iterative Prompt Refinement for Safer Text-to-Image Generation (2025.emnlp-main)
Copied to clipboard
| Challenge: | Existing safety methods for text-to-image models ignore the images produced . this can result in unsafe outputs or unnecessary changes to already safe prompts . |
| Approach: | They propose an iterative prompt refinement algorithm that uses Vision Language Models to analyze prompts and generated images. |
| Outcome: | The proposed method improves safety while maintaining user intent and reliability comparable to existing methods. |
C2KD: Cross-layer and Cross-head Knowledge Distillation for Small Language Model-based Recommendation (2025.findings-acl)
Copied to clipboard
| Challenge: | Large Language Models (LLMs) show promise but their size and high inference costs limit deployment on resource-constrained devices. |
| Approach: | They propose a framework to transfer task-relevant knowledge from two complementary dimensions to Large Language Models (LLMs) Large Language models (LLMS) have demonstrated great potential in sequential recommendation tasks . |
| Outcome: | Extensive experiments across diverse model families show that the proposed framework achieves competitive performance compared to LLMs. |
On the Step Length Confounding in LLM Reasoning Data Selection (2026.findings-acl)
Copied to clipboard
Bing Wang, Rui Miao, Chen Shen, Shaotian Yan, Kaiyuan Liu, Ximing Li, Xiaosong Yuan, Sinan Fan, Jun Zhang, Jieping Ye
| Challenge: | Existing pipelines generate long reasoning data from more capable Large Language Models (LLMs) and apply manually heuristic or naturalness-based selection methods to filter high-quality samples. |
| Approach: | They propose to use supervised fine-tuning to generate long reasoning data from more capable Large Language Models and apply manually heuristic or naturalness-based selection methods to filter high-quality samples. |
| Outcome: | Experiments on four LLMs and five evaluation benchmarks show that the proposed approach is effective in mitigating step length confounding problem. |
LoGU: Long-form Generation with Uncertainty Expressions (2025.acl-long)
Copied to clipboard
Ruihan Yang, Caiqi Zhang, Zhisong Zhang, Xinting Huang, Sen Yang, Nigel Collier, Dong Yu, Deqing Yang
| Challenge: | Large Language Models (LLMs) generate factually incorrect content, i.e., hallucinations, despite impressive performance. |
| Approach: | They propose a framework to enable models to express uncertainty when unsure . they propose atomic claims to refine uncertainty and refine it using supervised fine-tuning and direct preference optimization to enhance uncertainty expression. |
| Outcome: | The proposed framework significantly improves accuracy, reduces hallucinations, and maintains comprehensiveness of responses. |
A Modular Approach for Clinical SLMs Driven by Synthetic Data with Pre-Instruction Tuning, Model Merging, and Clinical-Tasks Alignment (2025.acl-long)
Copied to clipboard
Jean-Philippe Corbeil, Amin Dada, Jean-Michel Attendu, Asma Ben Abacha, Alessandro Sordoni, Lucas Caccia, Francois Beaulieu, Thomas Lin, Jens Kleesiek, Paul Vozila
| Challenge: | Large language models such as GPT-4 have limited their deployment in clinical settings . a novel framework for adapting SLMs into high-performing clinical models is needed . |
| Approach: | They propose a framework for adapting large language models into high-performing clinical models . they pre-instruct experts on relevant medical and clinical corpora and model merging . |
| Outcome: | The proposed framework outperforms the existing model on the CLUE+ benchmark on medical entities and radiology reports. |
Beyond Binary Preferences: Semi-Online Label-Free GRACE-KTO with Group-Wise Adaptive Calibration for High-Quality Long-Text Generation (2025.findings-emnlp)
Copied to clipboard
| Challenge: | Generating high-quality long-text remains challenging for Large Language Models (LLMs), as conventional supervised fine-tuning fails to ensure overall quality due to its teacher-forcing nature. |
| Approach: | They propose a semi-online framework that transforms KTO’s binary signals into dynamically calibrated intra-group rewards. |
| Outcome: | The proposed framework transforms binary signals into dynamically calibrated intra-group rewards. |
PhonoThink: Improving Large Language Models’ Reasoning on Chinese Phonological Ambiguities (2025.emnlp-main)
Copied to clipboard
| Challenge: | Effectively resolving phonological ambiguities is crucial for robust natural language processing, as these ambiguity are pervasive in tasks ranging from speech-to-text, spelling correction, to offensive language detection. |
| Approach: | They propose a framework to enhance LLMs’ phonological capability through a multiple-stage training approach. |
| Outcome: | The proposed framework enables the base model to achieve comparable performance to a much larger model. |
Embedding Domain Knowledge for Large Language Models via Reinforcement Learning from Augmented Generation (2025.emnlp-main)
Copied to clipboard
| Challenge: | Existing approaches to embed knowledge into large language models have some limitations . static nature of training data and lack of knowledge in domains create knowledge gaps . |
| Approach: | They propose a method that iteratively cycles between sampling generations and optimizing the model through calculated rewards. |
| Outcome: | The proposed method outperforms baseline approaches on medical, legal, astronomy, and current events datasets. |
3DS: Medical Domain Adaptation of LLMs via Decomposed Difficulty-based Data Selection (2025.emnlp-main)
Copied to clipboard
Hongxin Ding, Yue Fang, Runchuan Zhu, Xinke Jiang, Jinyang Zhang, Yongxin Xu, Weibin Liao, Xu Chu, Junfeng Zhao, Yasha Wang
| Challenge: | Effective domain adaptation typically involves supervised fine-tuning on carefully selected instruction-tuned data. |
| Approach: | They propose a model-centric data selection framework that aligns data selection with the model’s knowledge distribution to improve model performance. |
| Outcome: | The proposed framework outperforms existing methods by up to 2.97% accuracy in the healthcare domain. |
Community-Aware Assessment of Social Textual Engagement and Resonance: A Human-Centric Perspective on User-Generated Content Evaluation (2026.acl-long)
Copied to clipboard
| Challenge: | Traditional Video Quality Assessment (VQA) focuses on aesthetic fidelity and technical distortions. |
| Approach: | They propose a new task that evaluates whether a UGC item has positive community resonance based on multimodal attributes rather than visual quality alone. |
| Outcome: | The proposed task outperforms state-of-the-art baselines on CASTER-Bench . it provides interpretable and empathetic reasoning paths that align with real community feedback. |
VERITAS: Leveraging Vision Priors and Expert Fusion to Improve Multimodal Data (2025.emnlp-main)
Copied to clipboard
| Challenge: | supervised fine-tuning (SFT) data is critical for large multimodal models . current methods suffer from factual errors and hallucinations due to inadequate visual perception . |
| Approach: | They propose a pipeline that integrates vision priors and state-of-the-art LMMs with statistical methods to enhance SFT data quality. |
| Outcome: | The proposed pipeline outperforms existing models in text-rich reasoning tasks while enhancing reasoning efficiency. |
ReasonRank: Empowering Passage Ranking with Strong Reasoning Ability (2026.acl-long)
Copied to clipboard
| Challenge: | Existing rerankers perform poorly in complex ranking scenarios due to the scarcity of reasoning-intensive training data. |
| Approach: | They propose an automated reasoning-intensive training framework which generates high-quality training labels from training queries and passages. |
| Outcome: | The proposed model outperforms baselines significantly and achieves much lower latency than the pointwise reranker. |
Calibrating Language Models with Adaptive Temperature Scaling (2024.emnlp-main)
Copied to clipboard
| Challenge: | Large language models' confidence scores are degraded after fine-tuning with reinforcement learning from human feedback. |
| Approach: | They propose a post-hoc calibration method that predicts a temperature scaling parameter for each token prediction. |
| Outcome: | Adaptive temperature scaling improves calibration by over 10% compared to prior methods . RLHF fine-tuning improves model accuracy, but degradation is not significant . |
MT-R1-Zero: Advancing LLM-based Machine Translation via R1-Zero-like Reinforcement Learning (2025.findings-emnlp)
Copied to clipboard
| Challenge: | Large-scale reinforcement learning (RL) methods have proven effective in enhancing the reasoning abilities of large language models. |
| Approach: | They propose an open-source adaptation of the R1-Zero RL framework for machine translation (MT) their code is available at https://github.com/fzp0424/MT-R1-zero. |
| Outcome: | The proposed framework surpasses towerinstruct-7B-v0.2 on the english-chinese benchmark by 1.26 points. |
DRIFT: Transferring Reasoning Priors for Efficient MLLM Fine-Tuning (2026.findings-acl)
Copied to clipboard
Chao Huang, Zeliang Zhang, Jiang Liu, Ximeng Sun, Jialian Wu, Xiaodong Yu, Ze Wang, Chenliang Xu, Emad Barsoum, Zicheng Liu
| Challenge: | Multimodal large language models (MLLMs) have made rapid progress in perception and alignment, but their reasoning ability often lags behind strong text-only LLMs. |
| Approach: | They propose a method that transfers reasoning knowledge in the gradient space while preserving multimodal alignment. |
| Outcome: | Experiments on multimodal reasoning benchmarks show that DRIFT outperforms naive merging and standard SFT. |
MR. Judge: Multimodal Reasoner as a Judge (2025.emnlp-main)
Copied to clipboard
| Challenge: | Effective reward modeling is especially valuable in reinforcement learning (RLHF) . |
| Approach: | They propose a paradigm for empowering general-purpose MLLMs judges with strong reasoning capabilities by using multiple-choice problem models instead of directly assigning scores. |
| Outcome: | The proposed model surpasses GPT-4o on VL-RewardBench and improves performance on MM-Vet by up to 7.7%. |
Learning from Cognition: Enhancing RL Efficiency for LLM Reasoning via Hierarchical Metacognitive Decomposition and Refinement (2026.acl-long)
Copied to clipboard
Zexu Sun, Yongcheng Zeng, Erxue Min, Heyang Gao, Bokai Ji, Dugang Liu, Xing Tang, Xiuqiang He, Xu Chen
| Challenge: | Recent advances in Large Language Models have demonstrated notable inferential capacities via reinforcement learning (RL) however, “zero-RL” approaches relying on fixed prompt templates introduce substantial sampling inefficiencies for weak LLMs. |
| Approach: | They propose a hierarchical metacognitive RL framework that decomposes zero-accuracy problems into subproblems and prompts the policy to refine answers by referencing previous wrong solutions. |
| Outcome: | The proposed framework improves sample utilization and sample efficiency and accelerates convergence compared to baselines. |
Beyond Modality Collapse: Taming Guided Modality Entropy for Omni-modal Emotion Reasoning (2026.findings-acl)
Copied to clipboard
| Challenge: | EmoOmni is a data paradigm for omni-modal large language models that can be used for emotion reasoning. |
| Approach: | They propose a data paradigm that interleaves guided tokens into reasoning traces to enforce structured evidence extraction. |
| Outcome: | The proposed paradigm over-relys on a dominant modality while neglecting complementary cues. |
LLMs Can Achieve High-quality Simultaneous Machine Translation as Efficiently as Offline (2025.findings-acl)
Copied to clipboard
| Challenge: | Large language models perform well in offline machine translation when the complete source sentence is provided . however, in many real scenarios, the source tokens arrive in a streaming manner and simultaneous machine translation is required . |
| Approach: | They propose a new paradigm that includes constructing supervised fine-tuning data for simultaneous machine translation (SiMT) to achieve SiMT, source and target tokens are rearranged into interleaved sequences, separated by special tokens according to varying latency requirements. |
| Outcome: | The proposed approach achieves state-of-the-art performance across various SiMT benchmarks and evaluation metrics while maintaining efficient auto-regressive decoding. |
S2R: Teaching LLMs to Self-verify and Self-correct via Reinforcement Learning (2025.acl-long)
Copied to clipboard
| Challenge: | Existing approaches to incentivize LLMs’ deep thinking abilities require large-scale data or significant training efforts. |
| Approach: | They introduce an efficient framework that enhances LLM reasoning by teaching models to self-verify and self-correct during inference. |
| Outcome: | The proposed framework outperforms models trained on long-CoT distilled data with 3.1k initialization samples and achieves an accuracy improvement of 51.0% to 81.6%. |
Retrieval-Augmented Fine-Tuning With Preference Optimization For Visual Program Generation (2025.acl-long)
Copied to clipboard
| Challenge: | Visual programming languages (VPLs) allow users to create programs through graphical interfaces, which results in easier accessibility and widespread usage in various domains. |
| Approach: | They propose to train VPLs from user instructions using large language models . they propose to use retrieval-augmented fine-tuning to leverage repetitive use of subroutines . |
| Outcome: | The proposed method outperforms prompting-based methods for LD generation accuracy even with smaller backbone models. |
Towards Enhancing Coherence in Extractive Summarization: Dataset and Experiments with LLMs (2024.emnlp-main)
Copied to clipboard
| Challenge: | Existing methods for extractive summarization lack coherence, despite improvements . a human-annotated dataset is used to improve coherency of extractive summary . |
| Approach: | They propose to use human-annotated datasets to create coherent extractive summaries . they use supervised fine-tuning and natural language user feedback to enhance coherence . |
| Outcome: | The proposed dataset shows that LLMs can produce coherent summaries with human feedback. |
EXPO-SQL: Execution-based Clause-level Policy Optimization for Text-to-SQL (2026.findings-acl)
Copied to clipboard
| Challenge: | Existing RL methods assign query-level rewards to all clauses, treating correct and incorrect clauses equally. |
| Approach: | They propose a method which provides fine-grained supervision through clause-level rewards. |
| Outcome: | Experiments on widely-used Text-to-SQL benchmarks show that EXPO-SqL outperforms existing methods by fine-grained clause-level learning. |
PEToolLLM: Towards Personalized Tool Learning in Large Language Models (2025.findings-acl)
Copied to clipboard
| Challenge: | Existing tool learning studies focus on general-purpose tool-use capability, but ignore the importance of personalized tool-user preferences. |
| Approach: | They propose a framework to adapt Large Language Models to personalized tool learning task, which is trained through supervised fine-tuning and direct preference optimization. |
| Outcome: | Extensive experiments on PEToolBench show that the proposed framework outperforms existing LLMs in the personalized tool learning task. |
GeoRA: Geometry-Aware Low-Rank Adaptation for RLVR (2026.acl-long)
Copied to clipboard
| Challenge: | Existing parameter-efficient methods for RLVR face limitations . low-rank adaptation methods do not account for the distinct optimization dynamics . |
| Approach: | They propose a low-rank adaptation method tailored for RLVR that exploits the anisotropic structure of RL update subspace and extracts its principal directions via Singular Value Decomposition (SVD). |
| Outcome: | Experiments on large reasoning models show that GeoRA outperforms strong low-rank baselines across RLVR settings while showing stronger generalization and less forgetting on out-of-domain tasks. |
PIRB: A Comprehensive Benchmark of Polish Dense and Hybrid Text Retrieval Methods (2024.lrec-main)
Copied to clipboard
| Challenge: | PIRB is a framework for text information retrieval in Polish . existing and new datasets are evaluated to evaluate the performance of 41 models . |
| Approach: | They propose a framework for 41 text information retrieval tasks in Polish . they evaluate over 20 dense and sparse retrieval models and build sparser-dense hybrid retrievers . |
| Outcome: | The proposed framework outperforms the best available methods in 41 tasks for Polish . the proposed models outperformed the best solutions available to date . |
AAPO: Enhancing the Reasoning Capabilities of LLMs with Advantage Margin (2026.acl-long)
Copied to clipboard
| Challenge: | Reinforcement learning (RL) has emerged as an effective approach for enhancing the reasoning capabilities of large language models. |
| Approach: | They propose an algorithm that optimizes cross-entropy loss using advantages enhanced through a margin-based estimation scheme. |
| Outcome: | Experimental results show that AAPO improves group relative advantage estimation compared to other methods. |
Learning to Edit Knowledge via Instruction-based Chain-of-Thought Prompting (2026.acl-long)
Copied to clipboard
| Challenge: | Existing knowledge editing methods focus on structured fact triples, overlooking diverse unstructured forms of factual information. |
| Approach: | They propose a method that allows LLMs to edit knowledge via **Chain of Thoughts** reasoning. |
| Outcome: | The proposed method achieves strong generalization across six diverse knowledge editing scenarios with a single round of training on three open-source language models. |
ChipSeek: Optimizing Verilog Generation via EDA-Integrated Reinforcement Learning (2026.acl-long)
Copied to clipboard
Zhirong Chen, Kaiyan Chang, Zhuolin Li, Cangyuan Li, Xinyang He, Chujie Chen, Mengdi Wang, Haobo Xu, Yinhe Han, Huawei Li, Ying Wang
| Challenge: | Existing approaches to optimize Register-Transfer Level (RTL) code fail to simultaneously optimize functional correctness and hardware efficiency metrics such as Power, Performance, and Area (PPA). |
| Approach: | They propose a hierarchical reward based reinforcement learning framework that integrates direct feedback from EDA simulators and synthesis tools into a reward mechanism. |
| Outcome: | The proposed framework integrates direct feedback from EDA simulators and synthesis tools into a hierarchical reward based reinforcement learning framework. |
Beyond Overlap Metrics: Rewarding Reasoning and Preferences for Faithful Multi-Role Dialogue Summarization (2026.findings-acl)
Copied to clipboard
| Challenge: | Existing methods for multi-role dialogue summarization favor surface-level imitation of references rather than genuine gains in faithfulness or alignment with human preferences. |
| Approach: | They propose a framework that couples explicit cognitive-style reasoning with reward-based optimization for multi-role dialogue summarization. |
| Outcome: | The proposed framework matches strong baselines on ROUGE and BERTScore, while in-depth analysis on SAMSum shows clear gains in factual faithfulness and model-based preference alignment. |
Towards Safety Reasoning in LLMs: AI-agentic Deliberation for Policy-embedded CoT Data Creation (2025.findings-acl)
Copied to clipboard
Tharindu Kumarage, Ninareh Mehrabi, Anil Ramakrishna, Xinyan Zhao, Richard Zemel, Kai-Wei Chang, Aram Galstyan, Rahul Gupta, Charith Peris
| Challenge: | Safety reasoning paradigms require high-quality policy-embedded chain-of-thought datasets . generating such data through human annotations is prohibitively expensive . |
| Approach: | They propose AIDSAFE: Agentic Iterative Deliberation for Safety Reasoning . AIDS AFE leverages multi-agent deliberation to iteratively expand reasoning on safety policies . |
| Outcome: | The proposed model improves policy adherence and reasoning quality while maintaining acceptable utility and over-refusal accuracy. |
QDMR-based Planning-and-Solving Prompting for Complex Reasoning Tasks (2024.lrec-main)
Copied to clipboard
| Challenge: | Existing Plan-and-Solve prompting methods are difficult to implement for complex questions. |
| Approach: | They propose a plan-and-solve prompting method based on Question Decomposition Meaning Representation (QDMR) it allows LLM to generate a QDMR graph to represent problem-solving logic . |
| Outcome: | The proposed method can represent and execute the problem-solving logic of complex questions more accurately than existing methods. |
From Isolated Scoring to Collaborative Ranking: A Comparison-Native Framework for LLM-Based Paper Evaluation (2026.findings-acl)
Copied to clipboard
Pujun Zheng, Jiacheng Yao, Jinquan Zheng, Chenyang Gu, Guoxiu He, Jiawei Liu, Yong Huang, Tianrui Guo, Wei Lu
| Challenge: | Large language models (LLMs) are currently used to evaluate scientific papers by assigning an absolute score to each paper independently. |
| Approach: | They propose a comparison-native framework for paper evaluation that integrates comparison into both data construction and model learning. |
| Outcome: | The proposed framework achieves an average relative improvement of 21.8% over the strong baseline DeepReview-14B, while exhibiting robust generalization to five previously unseen datasets. |
Robust Adaptation of Large Multimodal Models for Retrieval Augmented Hateful Meme Detection (2025.emnlp-main)
Copied to clipboard
| Challenge: | Large Multimodal Models (LMMs) have shown promise in hateful meme detection, but they face limitations like sub-optimal performance and limited out-of-domain generalization capabilities. |
| Approach: | They propose a robust adaptation framework for hateful meme detection that enhances in-domain accuracy and cross-domain generalization while preserving the general vision-language capabilities of LMMs. |
| Outcome: | The proposed framework outperforms larger agentic systems in detecting hateful memes under adversarial attacks while maintaining the general vision-language capabilities of LMMs. |
TopXGen: Topic-Diverse Parallel Data Generation for Low-Resource Machine Translation (2025.findings-emnlp)
Copied to clipboard
| Challenge: | In-context learning and similarity search have been shown to improve LLMs' performance in machine translation, but they lag behind when dealing with low-resource languages. |
| Approach: | They propose a method that uses an LLM to generate topic-specific target-side data in the LRL. |
| Outcome: | The proposed approach boosts LLM translation performance during in-context learning and fine-tuning. |
Understand the Implication: Learning to Think for Pragmatic Understanding (2025.findings-acl)
Copied to clipboard
| Challenge: | Existing methods rely on annotated labels but overlook the reasoning process humans naturally use to interpret implicit meaning. |
| Approach: | They propose a dataset that includes explicit reasoning for both correct and incorrect interpretations and propose supervised fine-tuning to improve their performance. |
| Outcome: | The proposed dataset improves LLMs' pragmatic understanding by 11.12% across model families and 16.10% over label trained models. |
Dynamic Rewarding with Prompt Optimization Enables Tuning-free Self-Alignment of Language Models (2024.emnlp-main)
Copied to clipboard
| Challenge: | Empirical evaluations on eight recent LLMs reveal that DRPO significantly enhances alignment performance, enabling base models to outperform their SFT/RLHF-tuned counterparts. |
| Approach: | They propose a tuning-free approach to self-alignment called Dynamic Rewarding with Prompt Optimization (DRPO) it leverages a dynamic rewarding mechanism to identify and rectify alignment weaknesses . |
| Outcome: | The proposed approach outperforms existing methods and is highly adaptable to various alignment challenges. |
Empowering Math Problem Generation and Reasoning for Large Language Model via Synthetic Data based Continual Learning Framework (2025.emnlp-main)
Copied to clipboard
| Challenge: | Existing learning frameworks for large language models (LLMs) for math problem generation are limited and lack quality data. |
| Approach: | They propose a synthetic data based continual learning framework to improve LLMs ability for MPG and math reasoning. |
| Outcome: | The proposed framework improves performance on large language models and math reasoning using supervised fine-tuning, data synthesis and direct preference optimization. |
FOLIO: Natural Language Reasoning with First-Order Logic (2024.emnlp-main)
Copied to clipboard
Simeng Han, Hailey Schoelkopf, Yilun Zhao, Zhenting Qi, Martin Riddell, Wenfei Zhou, James Coady, David Peng, Yujie Qiao, Luke Benson, Lucy Sun, Alexander Wardle-Solano, Hannah Szabó, Ekaterina Zubova, Matthew Burtell, Jonathan Fan, Yixin Liu, Brian Wong, Malcolm Sailor, Ansong Ni, Linyong Nan, Jungo Kasai, Tao Yu, Rui Zhang, Alexander Fabbri, Wojciech Kryscinski, Semih Yavuz, Ye Liu, Xi Lin, Shafiq Joty, Yingbo Zhou, Caiming Xiong, Rex Ying, Arman Cohan, Dragomir Radev
| Challenge: | Existing benchmarks for logical reasoning in large language models lack language naturalness or limited complexity. |
| Approach: | They propose to use first-order logic annotations to evaluate logical reasoning capabilities of large language models. |
| Outcome: | The proposed dataset evaluates the FOL reasoning ability of supervised fine-tuning on medium-sized language models. |
Revisiting the Self-Consistency Challenges in Multi-Choice Question Formats for Large Language Model Evaluation (2024.lrec-main)
Copied to clipboard
| Challenge: | Multi-choice questions (MCQs) are a common method for assessing the world knowledge of large language models. |
| Approach: | They propose three knowledge-equivalent question variants to assess LLMs' world knowledge . they propose option position shuffle, option label replacement, and conversion to a True/False format . |
| Outcome: | The proposed questions are shuffle, label replacement, and True/False format. |
GroupRank: A Groupwise Paradigm for Effective and Efficient Passage Reranking with LLMs (2026.findings-acl)
Copied to clipboard
Meixiu Long, Duolin Sun, Dan Yang, Yihan Jiao, Lei Liu, Jiahai Wang, Binbin Hu, Yue Shen, Jie Feng, Zhehao Tan, Junjie Wang, Lianzhen Zhong, Jian Wang, Peng Wei, Jinjie Gu
| Challenge: | Existing rankers excel in lexical-matching scenarios, while they struggle with complex queries requiring deep reasoning. |
| Approach: | They propose a new paradigm that balances flexibility and context awareness to unlock the full potential of groupwise reranking. |
| Outcome: | The proposed approach achieves a state-of-the-art 65.2 NDCG@10 on BRIGHT and surpasses baselines by 2.1 points on R2MED while delivering a 6.4 inference speedup. |
Learning More from Less: Exploiting Counterfactuals for Data-Efficient Chart Understanding (2026.acl-long)
Copied to clipboard
Jianzhu Bao, Haozhen Zhang, Kuicai Dong, Bozhi Wu, Sarthak Ketanbhai Modi, Zi Pong Lim, Yon Shin Teo, Wenya Wang
| Challenge: | Chart understanding is a critical capability for vision-language models, serving as a cornerstone for automated data analysis, document understanding, and scientific research. |
| Approach: | They propose a chart-efficient training framework to enhance counterfactual sensitivity by code modification and a similarity-based data selection strategy. |
| Outcome: | The proposed framework achieves superior or comparable performance to strong chart-specific VLMs while using significantly less training data. |
Anatomy of Unlearning: The Dual Impact of Fact Salience and Model Fine-Tuning (2026.findings-acl)
Copied to clipboard
| Challenge: | Existing studies assume that all facts are equally forgettable . popular facts, frequent and widely distributed, may be more deeply embedded than rare ones, making them harder to erase. |
| Approach: | They propose a benchmark to evaluate how unlearning differs between pretrained and supervised fine-tuned models when fact popularity is taken into account. |
| Outcome: | The proposed model is compared with pretrained and SFT models on the forget data and shows that it performs better on both models. |
Understanding the Dark Side of LLMs’ Intrinsic Self-Correction (2025.acl-long)
Copied to clipboard
Qingjie Zhang, Di Wang, Haoting Qian, Yiming Li, Tianwei Zhang, Minlie Huang, Ke Xu, Hewu Li, Liu Yan, Han Qiu
| Challenge: | Recent studies show that LLMs’ intrinsic self-correction fails without oracle labels as feedback. |
| Approach: | They propose to use one simple task and three complex tasks with state-of-the-art LLMs like ChatGPT, Llama, and DeepSeek to interpret LLM's intrinsic self-correction. |
| Outcome: | The proposed methods reveal the dark side of LLMs’ intrinsic self-correction for different tasks, especially for those failure cases. |
GRAT: Guiding Retrieval-Augmented Reasoning through Process Rewards Tree Search (2025.acl-long)
Copied to clipboard
| Challenge: | Existing methods to enhance large models for multi-hop question-answering lack the ability for multipath exploration, strategic look-ahead, stepwise evaluation, and global selection. |
| Approach: | They propose an algorithm guided by Monte Carlo Tree Search and process rewards that assigns fine-grained rewards to each step in the search path. |
| Outcome: | The proposed algorithm outperforms various RAG-based methods on four multihop QA datasets and shows that it can self-train and self-update. |
TOWER+: Bridging Generality and Translation Specialization in Multilingual LLMs (2026.acl-long)
Copied to clipboard
Ricardo Rei, Nuno M Guerreiro, José Pombal, João Alves, Amin Farajian, Pedro Teixeirinha, Andre Martins
| Challenge: | Large Language Models (LLMs) are emerging as the de facto solution for multilingual machine translation. |
| Approach: | They propose a suite of LLMs that can be fine-tuned to deliver strong performance on translation and multilingual general-purpose text capabilities. |
| Outcome: | The proposed models outperform existing models on translation and general-purpose tasks. |
Learning from Near-Misses: Error-Aware Contrastive Few-Shot Learning for NL2Formula (2026.acl-long)
Copied to clipboard
| Challenge: | Existing spreadsheet formulas often produce near-miss outputs due to an incorrect function, operator, or reference. |
| Approach: | They propose an abstract syntax tree-based error taxonomy that organizes common error modes by the kind of decision that goes wrong in the parse tree. |
| Outcome: | The proposed framework improves Exact Match (EM) by 6.4 points over supervised fine-tuning and matches self-consistency (SC@5) accuracy. |
Does It Run and Is That Enough? Revisiting Text-to-Chart Generation with a Multi-Agent Approach (2025.findings-emnlp)
Copied to clipboard
| Challenge: | 15% of generated scripts fail to execute, even after supervised fine-tuning and reinforcement learning. |
| Approach: | They propose a lightweight multi-agent pipeline that separates drafting, execution, repair, and judgment . the system reduces execution errors to 4.5% within three repair iterations . |
| Outcome: | The proposed pipeline reduces execution errors to 4.5% within three repair iterations while requiring significantly less compute. |
LLM-based Rumor Detection via Influence Guided Sample Selection and Game-based Perspective Analysis (2025.acl-long)
Copied to clipboard
Zhiliang Tian, Jingyuan Huang, Zejiang He, Zhen Huang, Menglong Lu, Linbo Qiao, Songzhu Mei, Yijie Wang, Dongsheng Li
| Challenge: | Existing methods for rumor detection on social media are limited by limited modeling capacity and insufficient training corpora. |
| Approach: | They propose an SFT-based rumor detection model with Influence guided Sample selection and Game-based multi-perspective analysis to address these issues. |
| Outcome: | The proposed model outperforms existing SOTA on three datasets. |
Rationalize and Align: Enhancing Writing Assistance with Rationale via Self-Training for Improved Alignment (2025.findings-acl)
Copied to clipboard
| Challenge: | Existing writing assistants rely on supervised fine-tuning to optimize models for multiple revisions. |
| Approach: | They propose a framework that enhances WA performance with rationale and alignment. |
| Outcome: | The proposed framework outperforms state-of-the-art WAs and the closed-source GPT-4o by 3.9 and 7.1 points on average across eight well-established writing-related test sets. |
CogDual: Enhancing Dual Cognition of LLMs via Reinforcement Learning with Implicit Rule-Based Rewards (2025.emnlp-main)
Copied to clipboard
| Challenge: | Existing approaches to role-playing language models rely on prompt engineering or supervised fine-tuning to emulate character behaviors but neglect the underlying cognitive mechanisms driving these behaviors. |
| Approach: | They propose a novel RPLA adopting a cognize-then-respond reasoning paradigm that leverages dual cognition for more contextually grounded and psychologically coherent responses. |
| Outcome: | The proposed RPLA outperforms baselines and generalizes effectively across diverse role-playing tasks. |
MrGuard: A Multilingual Reasoning Guardrail for Universal LLM Safety (2025.emnlp-main)
Copied to clipboard
| Challenge: | Large Language Models (LLMs) are susceptible to adversarial attacks such as jailbreaking, which can elicit harmful or unsafe behaviors. |
| Approach: | They propose a multilingual guardrail with reasoning for prompt classification that integrates culturally and linguistically nuanced variants and supervised fine-tuning. |
| Outcome: | The proposed guardrail outperforms baselines across in-domain and out-of-domain languages by more than 15%. |
Improving the Language Understanding Capabilities of Large Language Models Using Reinforcement Learning (2025.findings-emnlp)
Copied to clipboard
| Challenge: | Instruction-fine-tuned large language models (LLMs) under 14B parameters underperform on NLU tasks . we explore a framework to improve the NLU capabilities of LLMs . |
| Approach: | They propose to use Proximal Policy Optimization to improve NLU capabilities . they frame NLU as a reinforcement learning environment and optimize for reward signals . |
| Outcome: | The proposed framework outperforms supervised fine-tuning on GLUE and superGLUE tasks. |
Why Supervised Fine-Tuning Fails to Learn: A Systematic Study of Incomplete Learning in Large Language Models (2026.acl-long)
Copied to clipboard
Chao Xue, Yao Wang, Mengqiao Liu, Di Liang, Xingsheng Han, Peiyang Liu, Xianjie Wu, Chenyao Lu, Lei Jiang, Yu Lu, Haibo Shi, Shuang Liang, Minlong Peng, Flora D. Salim
| Challenge: | Incomplete learning is widespread and heterogeneous in large language models . authors identify five recurrent sources of incomplete learning: missing prerequisite knowledge, conflicts between SFT supervision and pre-training knowledge, internal inconsistencies within SFT data, left-side forgetting during sequential fine-tuning, and insufficient optimization for rare or complex patterns. |
| Approach: | They propose a diagnostic-first framework that maps incomplete learning to causes . they identify five recurrent sources of incomplete learning: missing prerequisite knowledge, conflicts between supervision and pre-training knowledge, internal inconsistencies, left-side forgetting during sequential fine-tuning, and insufficient optimization for rare or complex patterns. |
| Outcome: | The proposed framework maps incomplete learning to causes using observable training and inference signals. |
End-to-End Optimization of LLM-Driven Multi-Agent Search Systems via Heterogeneous-Group-Based Reinforcement Learning (2026.acl-long)
Copied to clipboard
| Challenge: | Existing multi-agent reinforcement learning methods depend on large critic networks to evaluate joint actions, leading to instability and high memory costs. |
| Approach: | They propose a method to optimize large language models for agent-specific roles . they propose combining agent-based frameworks with retrieval-augmented generation . |
| Outcome: | Experiments show that multi-agent group policy optimization outperforms baselines in task performance and computational efficiency. |
Leveraging Loanword Constraints for Improving Machine Translation in a Low-Resource Multilingual Context (2025.emnlp-main)
Copied to clipboard
| Challenge: | a recent study addresses the challenge of adapting loanwords during the translation process in low-resource languages. |
| Approach: | They propose a method that augments source sentences with loanword constraints . they then integrate loanwords as external linguistic knowledge into machine translation systems . |
| Outcome: | The proposed approach improves translation quality and handling loanword adaptation correctly in target languages. |
𝒜3: Automatic Alignment Framework for Attributed Text Generation (2025.acl-long)
Copied to clipboard
| Challenge: | Existing approaches of aligning large language models to follow user instructions can lead to undue emphasis on irrelevant documents, which in turn reduces the quality of responses. |
| Approach: | They propose to use a framework to automatically generate high-quality attributed query-response pairs for both supervised fine-tuning and preference optimization stages without human annotation. |
| Outcome: | The proposed framework can generate high-quality attributed query-response pairs without human annotation without human intervention. |
Aspect-Oriented Summarization for Psychiatric Short-Term Readmission Prediction (2025.emnlp-main)
Copied to clipboard
WonJin Yoon, Boyu Ren, Spencer Thomas, Chanhwi Kim, Guergana K Savova, Mei-Hua Hall, Timothy A. Miller
| Challenge: | Recent advances in large language models have enabled the automated processing of lengthy documents even without supervised training on a task-specific dataset. |
| Approach: | They propose a method for processing the summaries of long documents using different aspect-oriented prompts and integrate the information signals from these different prompts for supervised training of transformer models. |
| Outcome: | The proposed method improves on a high-impact task predicting readmissions from a psychiatric discharge using real-world data from four hospitals. |
LLMs are Better Than You Think: Label-Guided In-Context Learning for Named Entity Recognition (2025.emnlp-main)
Copied to clipboard
| Challenge: | Named Entity Recognition (NER) tasks are performed using only a few demonstrations. |
| Approach: | They propose a method that leverages training labels through token-level statistics to improve ICL performance. |
| Outcome: | The proposed method outperforms existing methods on five NER datasets and is robust in low-resource settings. |
CARO: Chain-of-Analogy Reasoning Optimization for Robust Content Moderation (2026.findings-acl)
Copied to clipboard
| Challenge: | Current large language models struggle with ambiguous content moderation cases due to misleading "decision shortcuts" . authors propose a two-stage training framework to induce robust analogical reasoning in LLMs . |
| Approach: | They propose a two-stage training framework to induce robust analogical reasoning in LLMs . they bootstrap analogy reasoning chains via retrieval-augmented generation and SFT . |
| Outcome: | The proposed framework outperforms state-of-the-art reasoning models and specialized moderation models on ambiguous moderation benchmarks. |
GFT: From Imitation to Reward Fine-Tuning with Unbiased Group Advantages and Dynamic Coefficient Rectification (2026.findings-acl)
Copied to clipboard
| Challenge: | Existing studies have demonstrated that supervised fine-tuning and reinforcement learning are effective in integrating knowledge injection with robust generalization. |
| Approach: | They propose a unified post-training framework that addresses intrinsic limitations of supervised fine-tuning and reinforcement learning. |
| Outcome: | The proposed framework surpasses SFT-based methods and yields policies that integrate more smoothly with subsequent RL training. |
Compound AI Systems Optimization: A Survey of Methods, Challenges, and Future Directions (2025.emnlp-main)
Copied to clipboard
| Challenge: | Recent advances in large language models (LLMs) and AI systems have led to a paradigm shift in the design and optimization of complex workflows. |
| Approach: | They propose a systematic review of recent progress in optimizing compound AI systems . they formalize the notion of compound AI system optimization and classify existing methods along several key dimensions . |
| Outcome: | The proposed methods outperform existing methods in the field of compound AI and highlight open research challenges and future directions. |
Teaching Language Models to Check Grounded Claim Factuality with Human Test-Taking Strategies (2026.acl-long)
Copied to clipboard
| Challenge: | Existing methods for factuality checking require dataset-specific threshold tuning, while LLM-based approaches often use direct prompting. |
| Approach: | They propose to use a reading comprehension task to check for true/false claim factuality and prompt LLMs with explicit test-taking strategies for efficient reasoning. |
| Outcome: | The proposed method reduces token usage by over 80% compared to unguided open-ended reasoning and achieves competitive performance to more expensive alternatives. |
Are Large Language Models Reliable Reviewers? A Benchmark for Error Detection in Financial Documents (2026.findings-acl)
Copied to clipboard
Ying He, Zhouhong Gu, Zhecheng Hu, Yubo Zhou, Hao Shen, Jiaqing Liang, Zhaoqian Dai, Ma Shuguang, Fei Yu, Yanghua Xiao, Zhixu Li
| Challenge: | Existing LLMs struggle to identify errors in financial documents, a study shows . 18% of financial practitioners make errors daily, one-third make errors several times weekly, and 59% make errors multiple times monthly. |
| Approach: | They introduce FinED-Bench, a publicly available Benchmark for financial error detection . it covers nine real-world financial scenarios and includes over 900 documents in 2025 . supervised fine-tuning can significantly improve the performance of weaker LLMs, they show . |
| Outcome: | The proposed benchmark covers nine real-world financial scenarios and includes over 900 documents reported in 2025 that are unseen by existing language models. |
How Much Would a Clinician Edit This Draft? Evaluating LLM Alignment for Patient Message Response Drafting (2026.acl-long)
Copied to clipboard
Parker Seegmiller, Joseph Gatto, Sarah E. Greer, Ganza Belise Isingizwe, Rohan Ray, Timothy E. Burdick, Sarah Masud Preum
| Challenge: | Large language models (LLMs) have been shown to be effective in drafting patient portal responses, yet their integration into clinical workflows raises various concerns. |
| Approach: | They propose a taxonomy of thematic elements in clinician responses and a framework for assessing clinician editing load of LLM-drafted responses at both content and theme levels. |
| Outcome: | The proposed framework assesses the editing load of LLM-drafted responses at both content and theme levels. |
AscendKernelGen: LLM-Driven Kernel Generation for NPUs (2026.findings-acl)
Copied to clipboard
Xinzi Cao, Jianyang Zhai, Pengfei Li, Zhiheng Hu, Cen Yan, null Mubingxu, Guanghuan Fang, Bin She, Jiayu Li, Yihan Su, Dongyang Tao, Feidiao Yang, Chang-Dong Wang, Yutong Lu, Weicheng Xue, Bin Zhou, Yonghong Tian
| Challenge: | Neural Processing Units (NPUs) are critical for AI infrastructure, but their development remains a bottleneck due to vendor-specific Domain-Specific Languages (DSLs). |
| Approach: | They propose a framework for NPU kernel development that bridges the gap in hardware-specific coding . compiler success on complex Level-2 kernels improves from 0% to 95.5%, they say . |
| Outcome: | The proposed framework bridges the gap in hardware-specific coding, showing a near-zero success rate on complex kernels. |
ChemReason-Bench: Benchmarking Large Language Models for Procedural Reasoning in Experimental Chemistry (2026.acl-long)
Copied to clipboard
| Challenge: | Experimental protocols in organic synthesis specify not only the intended transformation, but also an executable sequence of operations and conditions. |
| Approach: | They propose a human-validated benchmark for verifiable experimental procedure reasoning . they instantiate 7306 benchmark tasks across six complementary formats . |
| Outcome: | The proposed benchmarks show that the evaluations are less diagnostic of procedure-level decision making. |
Tailored Primitive Initialization is the Secret Key to Reinforcement Learning (2026.acl-long)
Copied to clipboard
| Challenge: | Reinforcement learning (RL) has emerged as a powerful paradigm for improving the reasoning capabilities of large language models. |
| Approach: | They propose a pipeline that automatically discovers thinking token patterns with reasoning primitives and curates SFT datasets to prepare LLMs for RL. |
| Outcome: | The proposed pipeline outperforms baseline methods on mathematical and logical reasoning benchmarks on RL tasks. |
Magnet: Multi-turn Tool-use Data Synthesis and Distillation via Graph Translation (2025.acl-long)
Copied to clipboard
Fan Yin, Zifeng Wang, I-Hung Hsu, Jun Yan, Ke Jiang, Yanfei Chen, Jindong Gu, Long Le, Kai-Wei Chang, Chen-Yu Lee, Hamid Palangi, Tomas Pfister
| Challenge: | Large language models have been shown to be effective in multi-turn interactions . however, their performance may be limited in complex, multi-turned interactions involving users and multiple tools. |
| Approach: | They propose a framework for synthesizing high-quality training trajectories to enhance the function calling capability of large language model agents in multi-turn conversations with humans. |
| Outcome: | The proposed model outperforms the teacher model by 68.01 on BFCL-v3 and 73.30 on ToolQuery. |
MoRI: Learning Motivation-Grounded Reasoning for Scientific Ideation in Large Language Models (2026.acl-long)
Copied to clipboard
| Challenge: | Existing LLMs emulate human research workflows but lack scientific grounding . empirical results show that MoRI outperforms strong commercial LLM models . |
| Approach: | They propose a framework that explicitly learns scientific reasoning from research motivations to methodologies. |
| Outcome: | The proposed framework outperforms commercial LLMs and agentic baselines in novelty, technical rigor, and feasibility. |
VEHME: A Vision-Language Model For Evaluating Handwritten Mathematics Expressions (2025.emnlp-main)
Copied to clipboard
| Challenge: | VEHME is a vision language model for assessing handwritten math answers . traditional methods of assessing student work are limited by time constraints, class sizes and cognitive load . |
| Approach: | They propose a Vision-Language Model for Evaluating Handwritten Mathematics Expressions to assess handwritten math responses with high accuracy and interpretable reasoning traces. |
| Outcome: | VEHME achieves state-of-the-art performance among open-source models and approaches accuracy of proprietary systems. |
Text Detoxification: Data Efficiency, Semantic Preservation and Model Generalization (2025.emnlp-main)
Copied to clipboard
| Challenge: | Existing methods for detoxification of text often rely on manually annotated data . xiangli: "detoxification of texts is a powerful way to remove toxic content" |
| Approach: | They propose a reinforcement learning framework that optimizes detoxification and semantic preservation without annotating large amounts of data. |
| Outcome: | The proposed method overcomes major limitations and surpasses humanannotated references across multiple benchmarks. |
Training Language Models to Use Prolog as a Tool (2026.findings-acl)
Copied to clipboard
| Challenge: | Language models often produce plausible but incorrect reasoning traces that are difficult to verify. |
| Approach: | They train language models to use Prolog as an external symbolic reasoning tool . they find an accuracy–auditability trade-off between tuning for correctness alone and using Prolog only for the final computation . |
| Outcome: | The proposed model outperforms supervised fine-tuning on a clean version of GSM8K. |
ShopSimulator: Evaluating and Exploring RL-Driven LLM Agent for Shopping Assistants (2026.acl-long)
Copied to clipboard
Pei Wang, Yanan Wu, Xiaoshuai Song, Weixun Wang, Gengru Chen, Zhongwen Li, Kezhong Yan, Qi Liu, Ken Deng, Shuaibing Zhao, Shaopan Xiong, Xuepeng Liu, Xuefeng Chen, Wanxi Deng, Wenbo Su, Bo Zheng
| Challenge: | Existing studies on large language model-based agents focus on evaluation benchmarks without training support. |
| Approach: | They propose a large-scale Chinese shopping simulation environment that uses large language models to train agents. |
| Outcome: | The proposed model performs poorly in a large-scale and challenging shopping environment in China. |
TexOCR: Advancing Document OCR Models for Compilable Page-to-LaTeX Reconstruction (2026.acl-long)
Copied to clipboard
| Challenge: | Existing document OCR largely targets plain text or Markdown, discarding structural and executable properties that make LaTeX essential for scientific publishing. |
| Approach: | They propose a benchmark and a training corpus for document reconstruction . they train a 2B-parameter model using supervised fine-tuning and reinforcement learning . |
| Outcome: | The proposed model improves on existing models using supervised fine-tuning and reinforcement learning with verifiable rewards. |
Glance-or-Gaze: Incentivizing LMMs to Adaptively Focus Search via Reinforcement Learning (2026.findings-acl)
Copied to clipboard
| Challenge: | Existing search-augmented approaches rely on indiscriminate whole-image retrieval and lack deep iterative reflection, limiting their effectiveness on complex visual queries. |
| Approach: | They propose a fully autonomous framework that shifts from passive perception to active visual planning and introduces a Selective Gaze mechanism that dynamically chooses whether to glance at global context or gaze into high-value regions. |
| Outcome: | Experiments across six benchmarks demonstrate state-of-the-art performance. |
Glyph: Scaling Context Windows via Visual-Text Compression (2026.acl-long)
Copied to clipboard
Jiale Cheng, Yusen Liu, Xinyu Zhang, Yulin Fei, Wenyi Hong, Ruiliang Lyu, Weihan Wang, Zhe Su, Xiaotao Gu, Xiao Liu, Yushi Bai, Jie Tang, Hongning Wang, Minlie Huang
| Challenge: | Large language models (LLMs) traditionally represent text as sequences of discrete tokens . a long-context scaling problem requires processing more tokens more efficiently . |
| Approach: | They propose a framework that renders long texts into compact visual pages and processes them with a vision-language model. |
| Outcome: | The proposed framework renders long texts into compact visual pages and processes them with a vision-language model. |
SAFER: Advancing Safety Alignment via Efficient Ex-Ante Reasoning (2026.findings-acl)
Copied to clipboard
| Challenge: | Existing alignment methods struggle to cover diverse safety scenarios and remain vulnerable to adversarial attacks. |
| Approach: | They propose a framework for 'S**afety' alignment via e**F**ficient' E**x-Ante-R**easoning that instantiates structured Ex-Ance reasoning and embeds predefined safety rules. |
| Outcome: | The proposed framework enhances safety performance while maintaining usefulness and efficiency. |
Why Does Reinforcement Learning Generalize? A Feature-Level Mechanistic Study of Post-Training in Large Language Models (2026.acl-long)
Copied to clipboard
| Challenge: | Reinforcement learning (RL)-based post-training often improves the reasoning performance of large language models beyond the training domain, while supervised fine-tuning (SFT) frequently leads to general capabilities forgetting. |
| Approach: | They propose a feature-level mechanistic analysis methodology to probe RL generalization using a controlled experimental setup. |
| Outcome: | The proposed method identifies a compact, task-agnostic set of features that directly mediate generalization across diverse tasks. |
On-the-Fly VLA Adaptation via Test-Time Reinforcement Learning (2026.acl-long)
Copied to clipboard
Changyu Liu, Yiyang Liu, Taowen Wang, Qiao Zhuang, James Chenhao Liang, Wenhao Yang, Renjing Xu, Qifan Wang, Dongfang Liu, Cheng Han
| Challenge: | Existing vision-language-action models are unsuitable for simulated or physical-world deployments . current methods fail when confronted with inherent real-world dynamic variability. |
| Approach: | They propose a test-time reinforcement learning framework that enables on-the-fly policy adaptation during inference. |
| Outcome: | Empirical results show that the proposed framework improves adaptability, stability and task success in dynamic, previously unseen scenarios. |
Thinking in Schemas: Robust Syllogistic Reasoning in LLMs (2026.acl-long)
Copied to clipboard
| Challenge: | syllogistic reasoning models often mistake what sounds true for what is formally valid . content effect is a limitation of sluggish reasoning, which can lead to invalid conclusions . eisape et al., 2024: a key open problem for formal inference in natural language. |
| Approach: | They propose a schema-guided framework that disentangles semantic plausibility from logical validity. |
| Outcome: | The proposed framework outperforms existing frameworks while reducing CE. |
ReTRE: Benchmarking LLM Transfer Robustness with Structure-Preserving Variants (2026.acl-long)
Copied to clipboard
| Challenge: | Learning transfer theory emphasizes that applying acquired knowledge to novel manifestations is a key signal of deep understanding |
| Approach: | They propose a benchmark that probes transfer robustness along two rewrite levels: Near Transfer and Far Transfer. |
| Outcome: | The proposed benchmark demonstrates that large language models are robust when faced with novel manifestations of the same problem. |
Beyond Fully Random Masking: Attention-Guided Denoising and Optimization for Diffusion Language Models (2026.acl-long)
Copied to clipboard
| Challenge: | Existing methods for full-attention dLLMs rely on random masking strategies that overlook intrinsic token dependencies. |
| Approach: | They propose an attention-guided denoising and optimization framework that aligns training and optimization with attention-derived dependencies. |
| Outcome: | The proposed framework outperforms state-of-the-art methods on mathematical and coding benchmarks. |
Revealing the Seen, Imagining the Beyond: A Survey of Image-Grounded Chain-of-Thought Reasoning in Multimodal LLMs (2026.acl-long)
Copied to clipboard
| Challenge: | Recent advances in Multimodal Large Language Models (MLLMs) have shifted visual reasoning from tool-calling to end-to-end perceptionreasoning. |
| Approach: | They synthesize the emerging paradigm of Image-Grounded Chain-of-Thought (IG-CoT) they propose a method-centric taxonomy covering prompting, supervised fine-tuning, and reinforcement learning . |
| Outcome: | The proposed model is based on a method-centric taxonomy and benchmarks. |
Stable On-Policy Distillation through Adaptive Target Reformulation (2026.findings-acl)
Copied to clipboard
| Challenge: | Knowledge distillation (KD) is widely used for transferring capabilities from proprietary models to efficient open-source counterparts. |
| Approach: | They propose a method that constructs a geometric target distribution in logit space to emphasize agreement between the teacher and the student. |
| Outcome: | Experiments show that the proposed method outperforms supervised fine-tuning and existing on-policy baselines. |
Thinking-Based Non-Thinking: Solving the Reward Hacking Problem in Training Hybrid Reasoning Models via Reinforcement Learning (2026.acl-long)
Copied to clipboard
Siyuan Gan, Jiaheng Liu, Boyan Wang, Tianpei Yang, Runqing Miao, Yuyao Zhang, Fanyu Meng, Junlan Feng, Linjian Meng, Jing Huo, Yang Gao
| Challenge: | Existing work on large reasoning models (LRMs) focuses on using reinforcement learning (RL) to train hybrid reasoning models that automatically decide whether to engage in thinking or not based on the complexity of the query. |
| Approach: | They propose to use reinforcement learning to train hybrid reasoning models that automatically decide whether to engage in thinking or not based on the complexity of the query. |
| Outcome: | The proposed model reduces token usage by around 50%$ compared to DeepSeek-R1-Distill-Qwen-1.5B/7B and DeepScaleR-1.5b, while significantly improving accuracy. |
DPEPO: Diverse Parallel Exploration Policy Optimization for LLM-based Agents (2026.acl-long)
Copied to clipboard
JunShuo Zhang, Chengrui Huang, Feng Guo, Zihan Li, Ke Shi, Menghua Jiang, Jiguo Yu, Shuo Shang, Shen Gao
| Challenge: | Existing approaches to large language model (LLM) agents that follow the sequential "reason-then-act" paradigm suffer from limited exploration and incomplete environmental understanding as they interact with only a single environment per step. |
| Approach: | They propose a paradigm that enables an agent to interact with multiple environments simultaneously and share cross-trajectory experiences. |
| Outcome: | The proposed paradigm achieves state-of-the-art (SOTA) success rates while maintaining comparable efficiency to strong sequential baselines. |