Papers with training efficiency
Advancing African-Accented English Speech Recognition: Epistemic Uncertainty-Driven Data Selection for Generalizable ASR Models (2025.acl-srw)
Copied to clipboard
| Challenge: | Accents play a pivotal role in shaping human communication, a new study finds . existing ASR systems often perform inadequately, even mispronouncing African names . |
| Approach: | They propose a method that uses epistemic uncertainty to automate annotation to reduce costs and human labor. |
| Outcome: | The proposed method reduces costs and human labor by reducing data annotation and epistemic uncertainty. |
EffEval: A Comprehensive Evaluation of Efficiency for MT Evaluation Metrics (2023.findings-emnlp)
Copied to clipboard
| Challenge: | a recent surge of interest in developing evaluation metrics based on pretrained large language models (LLMs) can better cope with lexical variation. |
| Approach: | They propose to replace computation-intensive transformers with lighter alternatives and employ linear and quadratic approximations for alignment algorithms on top of LLM representations. |
| Outcome: | The proposed approach replaces computation-intensive transformers with lighter alternatives and employs linear and quadratic approximations for alignment algorithms on top of LLM representations. |
Rationale-Guided Distillation for E-Commerce Relevance Classification: Bridging Large Language Models and Lightweight Cross-Encoders (2025.coling-industry)
Copied to clipboard
| Challenge: | Large-scale e-commerce search systems typically follow a multi-step process to retrieve relevant products for a given query. |
| Approach: | They propose a distillation approach that uses "rationales" generated by Large Language Models to guide smaller cross-encoder models. |
| Outcome: | The proposed model achieves ROC-AUC improvements of 1.4% on 9 multilingual e-commerce datasets, 2.4% on 3 ESCI datasets and 6% on GLUE datasets while being 50 times faster per sample. |
Improving Mongolian-Chinese Neural Machine Translation with Morphological Noise (P19-2)
Copied to clipboard
| Challenge: | Existing models for Mongolian-Chinese translation are based on recurrent, convolutional neural networks or completely eliminate recurrence connections. |
| Approach: | They propose a adversarial training model to alleviate the UNK problem in Mongolian-Chinese machine translation by adding a screener to the model to emphasize the added Mongolian morphological noise. |
| Outcome: | The proposed model reduces training time and improves accuracy in Mongolian-Chinese translation tasks. |
CytonMT: an Efficient Neural Machine Translation Open-source Toolkit Implemented in C++ (D18-2)
Copied to clipboard
| Challenge: | Neural machine translation (NMT) has made remarkable progress over the past few years. |
| Approach: | They propose to use C++ and NVIDIA’s GPU-accelerated libraries to build an open-source neural machine translation toolkit called CytonMT. |
| Outcome: | The proposed toolkit accelerates the training speed by 64.5% to 110.8% on neural networks of various sizes, and achieves competitive translation quality. |
End-to-End Optimization for Multimodal Retrieval-Augmented Generation via Reward Backpropagation (2025.findings-emnlp)
Copied to clipboard
| Challenge: | MM-RAG is a promising approach for enhancing the reliability and factuality of large vision-language models . current methods focus on component-level optimizations and necessitate extensive component-specific training datasets . |
| Approach: | They propose a new paradigm that backpropagates global rewards to each component . this backpropage transforms local losses into specific local losses . |
| Outcome: | The proposed paradigm achieves high training efficiency on knowledge-intensive multimodal benchmarks. |
MoFE: Mixture of Frozen Experts Architecture (2025.naacl-industry)
Copied to clipboard
| Challenge: | Large Language Models (LLMs) are characterized by their immense size, often consisting of at least one billion parameters. |
| Approach: | They propose a mixture of Frozen Experts architecture that integrates PEFT and MoE to enhance both training efficiency and model scalability. |
| Outcome: | The proposed architecture outperforms other methods while achieving the highest efficiency. |
Multimodal Generation with Consistency Transferring (2025.findings-naacl)
Copied to clipboard
| Challenge: | Existing methods for multimodal content generation are limited to unimodal content production due to high training complexity, significant costs, and inadequate emphasis on model constraints. |
| Approach: | They propose a method to generate multimodal content with constraints on adjacent steps and a layer-based layer-constrained transfer between adjacent steps to improve denoising capabilities. |
| Outcome: | The proposed method improves the model’s ability to capture actions and depict backgrounds more effectively and improves video generation speed by approximately 40% and quality by about 39.3%. |
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. |
PILLOW: Enhancing Efficient Instruction Fine-tuning via Prompt Matching (2023.emnlp-industry)
Copied to clipboard
| Challenge: | Low-Rank Adaptation (LoRA) has been used to adapt Large Language Models to a variety of tasks, but it requires substantial computational resources to perform. |
| Approach: | They propose a low-rank adaptive learning approach that leverages LoRA's in-context learning capability through prompt matching via reinforcement learning in resource-constrained environments. |
| Outcome: | The proposed model improves LoRA performance on evaluation metrics and utilises consumer-grade GPU resources. |
Let the Expert Stick to His Last: Expert-Specialized Fine-Tuning for Sparse Architectural Large Language Models (2024.emnlp-main)
Copied to clipboard
| Challenge: | Existing studies on parameter-efficient fine-tuning (PEFT) for dense-architecture LLMs are lacking. |
| Approach: | They propose an expert-specialized fine-tuning method that tunes the experts most relevant to downstream tasks while freezing the other experts. |
| Outcome: | The proposed method matches or surpasses full-parameter fine-tuning. |
CoLLiE: Collaborative Training of Large Language Models in an Efficient Way (2023.emnlp-demo)
Copied to clipboard
Kai Lv, Shuo Zhang, Tianle Gu, Shuhao Xing, Jiawei Hong, Keyu Chen, Xiaoran Liu, Yuqing Yang, Honglin Guo, Tengxiao Liu, Yu Sun, Qipeng Guo, Hang Yan, Xipeng Qiu
| Challenge: | Large language models (LLMs) are increasingly pivotal in a wide range of tasks . however, the resources required for training these models necessitate efficient solutions . |
| Approach: | They propose a library that facilitates collaborative training of large language models . they use 3D parallelism, parameter-efficient fine-tuning methods and optimizers . |
| Outcome: | The proposed library has proven superior training efficiency in comparison with prevalent solutions in pre-training and fine-tuning scenarios. |
OpenRLHF: A Ray-based Easy-to-use, Scalable and High-performance RLHF Framework (2025.emnlp-demos)
Copied to clipboard
Jian Hu, Xibin Wu, Wei Shen, Jason Klein Liu, Weixun Wang, Songlin Jiang, Haoran Wang, Hao Chen, Bin Chen, Wenkai Fang, null Xianyu, Yu Cao, Haotian Xu, Yiming Liu
| Challenge: | Existing RLHF frameworks face inference bottlenecks and complexity barriers restricting their accessibility for newcomers. |
| Approach: | They propose an open-source RLHF framework that can be used to train large language models. |
| Outcome: | The proposed framework achieves superior training efficiency with speedups ranging from 1.22 to 1.68 across different model sizes compared to state-of-the-art frameworks, while requiring significantly fewer lines of code for implementation. |
PLATO-Ad: A Unified Advertisement Text Generation Framework with Multi-Task Prompt Learning (2022.emnlp-industry)
Copied to clipboard
Zeyang Lei, Chao Zhang, Xinchao Xu, Wenquan Wu, Zheng-yu Niu, Hua Wu, Haifeng Wang, Yi Yang, Shuanglong Li
| Challenge: | Online advertisement text generation models have achieved remarkable success in generating high-quality text ads, but some challenges remain, such as low-resource scenarios and training efficiency for multiple ad tasks. |
| Approach: | They propose a unified text ad generation framework with multi-task prompt learning to tackle low-resource ade generation problem and a multi-step prompt learning mechanism to efficiently solve multiple aed generation tasks. |
| Outcome: | The proposed framework outperforms the state-of-the-art on offline and online metrics. |
Aleph-Alpha-GermanWeb: Improving German-language LLM pre-training with model-based data curation and synthetic data generation (2026.eacl-long)
Copied to clipboard
Thomas F Burns, Letitia Parcalabescu, Stephan Waeldchen, Michael Barlow, Gregor Ziegltrum, Volker Stampa, Bastian Harren, Björn Deiseroth
| Challenge: | Recent studies show that data quality can significantly boost performance and training efficiency for large language models. |
| Approach: | They propose a German-language dataset curation pipeline that combines heuristic and model-based filtering techniques with synthetic data generation. |
| Outcome: | The proposed pipeline can be used to create a large-scale German pre-training dataset using common Crawl web data, fineweb2 and synthetically generated data conditioned on real, organic web data. |
GRAM: Fast Fine-tuning of Pre-trained Language Models for Content-based Collaborative Filtering (2022.naacl-main)
Copied to clipboard
| Challenge: | Content-based collaborative filtering (CF) predicts user-item interactions based on both items’ interaction history and item content information. |
| Approach: | They propose to combine item encodings with a multi-modality approach to improve training efficiency by 146x . |
| Outcome: | The proposed model improves training efficiency (up to 146x) on five datasets from two task domains of Knowledge Tracing and News Recommendation. |
Untie the Knots: An Efficient Data Augmentation Strategy for Long-Context Pre-Training in Language Models (2025.acl-long)
Copied to clipboard
| Challenge: | Large language models (LLMs) have prioritized expanding the context window from which they can incorporate more information. |
| Approach: | They propose a data augmentation strategy to enable large language models to gain long-context capabilities without the need to modify existing data mixture. |
| Outcome: | The proposed model outperforms existing models on 20 billion tokens and achieves 75% and 84.5% accuracy on RULER at 128K context length. |
Efficient Learning of Multiple NLP Tasks via Collective Weight Factorization on BERT (2022.findings-naacl)
Copied to clipboard
| Challenge: | Existing methods to fine-tune a model for multiple tasks require a large amount of memory and computing power. |
| Approach: | They propose to factorize the weighs of a pre-trained Transformer model to improve training efficiency across multiple tasks by using BERT-Large as an instantiation of the Transformer and the GLUE as the evaluation benchmark. |
| Outcome: | The proposed method matches or improves the original fine-tuned model’s performance for each task while effectively decreasing parameter requirements by two orders of magnitude. |
WISCA: A Lightweight Model Transition Method to Improve LLM Training via Weight Scaling (2026.findings-acl)
Copied to clipboard
Jiacheng Li, Jianchao Tan, Zhidong Yang, Pingwei Sun, Feiye Huo, Jiayu Qin, Xiangyu Zhang, Maoxin He, Guangming Tan, Weile Jia, Xunliang Cai, Tong Zhao
| Challenge: | Recent advances in training optimization for Transformer-based large language models lack systematic optimization of weight patterns during training. |
| Approach: | They propose a Weight Scaling method that rescales weights while preserving model outputs to improve model training efficiency and model quality. |
| Outcome: | The proposed method significantly improves convergence quality and loss reduction in LLMs with Grouped Query Attention architectures and LoRA fine-tuning tasks. |
Random Smooth-based Certified Defense against Text Adversarial Attack (2024.findings-eacl)
Copied to clipboard
| Challenge: | Textual adversarial examples train models on the worst-case text generated by substituting words in original texts with synonyms, but due to the discrete word embedding representations, the large search space hinders the robust training efficiency. |
| Approach: | They propose to treat the word substitution as a continuous perturbation on the word embedding representation and apply random smooth techniques to approximate the word replacement operation. |
| Outcome: | The proposed method outperforms conventional methods and improves the robustness in training. |
FTFT: Efficient and Robust Fine-Tuning by Transferring Training Dynamics (2025.coling-main)
Copied to clipboard
| Challenge: | Despite the success of fine-tuning Pre-trained Language Models, they remain susceptible to out-of-distribution input. |
| Approach: | They propose a novel approach that fine-tunes Pre-trained Language Models by transFerring Training dynamics (FTFT) FTFT uses more efficient reference models and aggressive early stopping . |
| Outcome: | The proposed approach improves the robustness of fine-tuned PLMs while reducing training costs. |
Selecting Informative Contexts Improves Language Model Fine-tuning (2021.acl-long)
Copied to clipboard
| Challenge: | Language model fine-tuning is computationally expensive and time-consuming . however, the inclusion of training examples that negatively affect performance is limited . |
| Approach: | They propose a general fine-tuning method that incorporates information gain filtration . they propose to release pre-trained secondary learners on common corpora to promote efficient fine-uning. |
| Outcome: | The proposed method achieves a median perplexity of 54.0 on a books dataset compared to 57.3 for standard fine-tuning. |
MEFT: Memory-Efficient Fine-Tuning through Sparse Adapter (2024.acl-long)
Copied to clipboard
| Challenge: | Parameter-Efficient Fine-tuning (PEFT) methods are limited on knowledge-intensive tasks due to the limited number of trainable parameters. |
| Approach: | They propose a mechanism that fine-tunes Large Language Models with larger adapters . they store and update the parameters of larger adapter adapters on the CPU . |
| Outcome: | The proposed method achieves comparable results to those obtained with larger memory capacities over the limited bandwidth of PCI Express (PCIe). |
LTRS: Improving Word Sense Disambiguation via Learning to Rank Senses (2025.coling-main)
Copied to clipboard
| Challenge: | Conventional training strategies only consider predefined senses for target words and learn each of them from relatively limited instances, neglecting the influence of similar ones. |
| Approach: | They propose a method to rank senses to improve the task of word Sense Disambiguation (WSD) by ranking an expanded list of sense definitions. |
| Outcome: | The proposed method achieves a SOTA F1 score of 79.6% in Chinese WSD and shows faster convergence than previous methods. |
SAPIENT: Mastering Multi-turn Conversational Recommendation with Strategic Planning and Monte Carlo Tree Search (2025.naacl-long)
Copied to clipboard
| Challenge: | Existing methods train RL-based agents with greedy action selection or sampling strategy and suffer from suboptimal conversational planning. |
| Approach: | They propose a Monte Carlo Tree Search-based CRS framework called SAPIENT . it consists of a conversational agent and a communication planner . |
| Outcome: | The proposed framework outperforms the state-of-the-art methods on four benchmark datasets. |
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. |
Influence Scores at Scale for Efficient Language Data Sampling (2023.emnlp-main)
Copied to clipboard
| Challenge: | Recent studies have shown that ML models can be fine-tuned on as much data as possible without degradation in performance metrics. |
| Approach: | They evaluate the applicability of influence scores in language classification tasks by random sampling and stress-testing one of the scores. |
| Outcome: | The proposed model can be fine-tuned on 50% of the original data without degradation in performance metrics. |
MoDULA: Mixture of Domain-Specific and Universal LoRA for Multi-Task Learning (2024.emnlp-main)
Copied to clipboard
Yufei Ma, Zihan Liang, Huangyu Dai, Ben Chen, Dehong Gao, Zhuoran Ran, Wang Zihan, Linbo Jin, Wen Jiang, Guannan Zhang, Xiaoyan Cai, Libin Yang
| Challenge: | Recent advances in open-source Large Language Models (LLMs) have achieved notable successes in natural language processing. |
| Approach: | They propose a Parameter Efficient Fine-Tuning paradigm for improved fine-tuning and parameter efficiency in multi-task learning. |
| Outcome: | The proposed model outperforms existing methods on multi-task learning while reducing training costs by over 80% without losing general capability. |
HiCL: Hierarchical Contrastive Learning of Unsupervised Sentence Embeddings (2023.findings-emnlp)
Copied to clipboard
| Challenge: | Existing methods that encode a sequence in its entirety for contrast with others often neglect local representation learning. |
| Approach: | They propose a hierarchical contrastive learning framework, HiCL, which considers local segment-level and global sequence-level relationships to improve training efficiency and effectiveness. |
| Outcome: | The proposed framework improves training efficiency and effectiveness by dividing a sequence into several segments and using local and global contrastive learning to model relationships. |
Efficient CTC Regularization via Coarse Labels for End-to-End Speech Translation (2023.eacl-main)
Copied to clipboard
| Challenge: | Developing techniques to support end-to-end speech translation is non-trivial because of the speech-text modality gap. |
| Approach: | They propose a coarse labeling approach that merges vocabulary labels via simple heuristic rules . they propose to use 256-bit truncation, division or modulo operations to regularize the encoder . |
| Outcome: | The proposed method can increase training efficiency while delivering better performance. |
LoRETTA: Low-Rank Economic Tensor-Train Adaptation for Ultra-Low-Parameter Fine-Tuning of Large Language Models (2024.naacl-long)
Copied to clipboard
| Challenge: | Existing methods for parameter-efficient fine-tuning are limited by the growing number of trainable parameters with the rapid deployment of Large Language Models (LLMs). |
| Approach: | They propose a parameter-efficient framework that reduces trainable parameters through tensor-train decomposition. |
| Outcome: | The proposed methods achieve comparable or better performance than most widely used methods with up to 100 fewer parameters on the LLaMA-2-7B models. |
PVPO: Pre-Estimated Value-Based Policy Optimization for Agentic Reasoning (2026.findings-acl)
Copied to clipboard
| Challenge: | extending grouping-based methods to agentic reasoning presents unique challenges . frequent environment interactions and tool invocations render intra-group advantage estimation unstable . |
| Approach: | They propose a grouping-based method that uses a single round of rollouts to stabilize advantage estimation. |
| Outcome: | a new RL framework outperforms grouping-based methods in retrieval tasks and advanced mathematical reasoning benchmarks. |
Curriculum Masking in Vision-Language Pretraining to Maximize Cross Modal Interaction (2024.naacl-long)
Copied to clipboard
| Challenge: | masked language modeling is widely used as a pretraining component in Vision and language (V+L) but performance on benchmarks has not received the attention it deserves. |
| Approach: | They propose a curriculum masking scheme that uses a parallel mask selection agent to mask tokens at a frequency proportional to the level of cross modal interaction necessary to reconstruct them. |
| Outcome: | The proposed method improves relational understanding on a wide range of V+L tasks. |
AVA: Attentive VLM Agent for Mastering StarCraft II (2026.findings-acl)
Copied to clipboard
| Challenge: | Existing StarCraft II benchmarks rely on abstract state representations that deviate from human perception . Existing systems rely only on abstract representations, creating an artificial gap between how humans process battlefield information and limiting ecological validity of learned behaviors. |
| Approach: | They introduce AVACraft, the first multimodal benchmark environment for complex decision-making in StarCraft II. |
| Outcome: | The AVACraft benchmark supports both traditional and modern multi-agent reinforcement learning paradigms. |
DIDS: Domain Impact-aware Data Sampling for Large Language Model Training (2025.emnlp-main)
Copied to clipboard
Weijie Shi, Jipeng Zhang, Yaguang Wu, Jingzhi Fang, Shibo Zhang, Yao Zhao, Hao Chen, Ruiyuan Zhang, Yue Cui, Jia Zhu, Sirui Han, Jiajie Xu, Xiaofang Zhou
| Challenge: | Existing approaches for optimizing domain-level sampling strategies struggle with maintaining intra-domain consistency and accurately measuring domain impact. |
| Approach: | They propose to use a Fisher-Information Matrix-guided metric to measure domain impact to ensure intra-domain consistency and accuracy. |
| Outcome: | The proposed model achieves 3.4% higher average performance while maintaining comparable training efficiency. |
CSMCIR: CoT-Enhanced Symmetric Alignment with Memory Bank for Composed Image Retrieval (2026.findings-acl)
Copied to clipboard
Zhipeng Qian, Zihan Liang, Yufei Ma, Ben Chen, Huangyu Dai, Yiwei Ma, Jiayi Ji, Chenyi Lei, Han Li, Xiaoshuai Sun
| Challenge: | Existing approaches to search for images using single-modality are limited by representation space fragmentation. |
| Approach: | They propose a unified representation framework that achieves efficient query-target alignment . they introduce a multi-level Chain-of-Thought prompting strategy that guides MLMs to generate discriminative, semantically compatible captions for target images . |
| Outcome: | The proposed framework achieves efficient query-target alignment through synergistic components. |
Adaptive Gating in Mixture-of-Experts based Language Models (2023.emnlp-main)
Copied to clipboard
| Challenge: | Existing models employ a fixed gating network where each token is computed by the same number of experts. |
| Approach: | They propose a flexible training strategy that allows tokens to be processed by a variable number of experts based on expert probability distribution. |
| Outcome: | The proposed model reduces training time and inference quality while maintaining sparsity while maintaining inference accuracy. |
SoftDedup: an Efficient Data Reweighting Method for Speeding Up Language Model Pre-training (2024.acl-long)
Copied to clipboard
| Challenge: | Current methods focus on detecting and removing duplicates, which risks the loss of valuable information and neglects the varying degrees of duplication. |
| Approach: | They propose a method that maintains dataset integrity while selectively reducing the sampling weight of data with high commonness. |
| Outcome: | The proposed method significantly improves training efficiency on deduplicated datasets and improves downstream accuracy by 1.77%. |
RLPrompt: Optimizing Discrete Text Prompts with Reinforcement Learning (2022.emnlp-main)
Copied to clipboard
Mingkai Deng, Jianyu Wang, Cheng-Ping Hsieh, Yihan Wang, Han Guo, Tianmin Shu, Meng Song, Eric Xing, Zhiting Hu
| Challenge: | Existing methods for finding the optimal prompt for a task are difficult to optimize. |
| Approach: | They propose an efficient discrete prompt optimization approach with reinforcement learning that generates the optimal discrete stimulus after training with reward. |
| Outcome: | The proposed approach is based on a parameter-efficient policy network that generates the optimal discrete prompt after training with reward. |
CoLA: Compute-Efficient Pre-Training of LLMs via Low-Rank Activation (2025.emnlp-main)
Copied to clipboard
Ziyue Liu, Ruijie Zhang, Zhengyang Wang, Mingsong Yan, Zi Yang, Paul D. Hovland, Bogdan Nicolae, Franck Cappello, Sui Tang, Zheng Zhang
| Challenge: | Large foundation models have become huge, but they consume computational resources in pretraining. |
| Approach: | They propose to replace full-size layers with compute-efficient auto-encoders that enforce low-rank activations throughout training. |
| Outcome: | The proposed method reduces the computing cost by 2pmbtimes and improves training throughput by 1.86pmtime. |
SiMFy: A Simple Yet Effective Approach for Temporal Knowledge Graph Reasoning (2023.findings-emnlp)
Copied to clipboard
| Challenge: | Existing models for temporal knowledge graph reasoning suffer from low training efficiency and insufficient generalization ability. |
| Approach: | They propose a temporal knowledge graph reasoning approach that uses multilayer perceptron to model the structural dependencies of events and adopts a fixed-frequency strategy to incorporate historical frequency during inference. |
| Outcome: | The proposed model achieves state-of-the-art performance with faster convergence speed and better generalization ability. |
K-order Ranking Preference Optimization for Large Language Models (2025.findings-acl)
Copied to clipboard
| Challenge: | Existing list-wise methods focus on optimizing list ranking consistency for LLMs to improve ranking abilities. |
| Approach: | They propose to extend the Plackett-Luce model to accommodate top-K ranking by extending the DPO’s Plact-Lucer model to dynamically determine appropriate K for different samples. |
| Outcome: | The proposed model can be extended to accommodate top-K ranking and improve training efficiency. |
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. |
Parameter-Efficient Fine-Tuning of Large Language Models via Deconvolution in Subspace (2025.coling-main)
Copied to clipboard
| Challenge: | Existing methods for parameter-efficient fine-tuning have been proposed to reduce time and resource costs. |
| Approach: | They propose a parameter-efficient fine-tuning method that combines the knowledge completion capability of deconvolution with the subspace learning ability, reducing the number of parameters required for fine-uning by 8 times. |
| Outcome: | The proposed method reduces the number of parameters required for fine-tuning by 8 times and achieves comparable or superior performance compared to existing models. |
Critic-Guided Decoding for Controlled Text Generation (2023.findings-acl)
Copied to clipboard
| Challenge: | Recent work has demonstrated reinforcement learning and weighted decoding as effective approaches to achieve a higher level of language control and quality with pros and cons. |
| Approach: | They propose a method that combines reinforcement learning and weighted decoding to train a critic from reward models. |
| Outcome: | The proposed method generates more coherent and well-controlled texts than previous methods on three controlled generation tasks, topic control, sentiment control, and detoxification. |
Self-Supervised Curriculum Learning for Spelling Error Correction (2021.emnlp-main)
Copied to clipboard
| Challenge: | Current approaches to SEC typically leverage a pre-training then fine-tuning procedure that treats data equally. |
| Approach: | They propose a self-supervised curriculum learning approach to improve model performance and model learning. |
| Outcome: | The proposed approach improves the model training and improves CL measurement. |
Robust Preference Optimization via Dynamic Target Margins (2025.findings-acl)
Copied to clipboard
| Challenge: | Direct Preference Optimization (DPO) is an efficient method for ensuring safety and reliability in practical applications. |
| Approach: | They propose a dynamic target margin preference optimization algorithm that adjusts reward margins at the pairwise level. |
| Outcome: | The proposed method achieves an average 4.4% improvement over baselines, setting new benchmarks for state-of-the-art performance. |
Federated LoRA Fine-Tuning with Pipelined Error-Mitigated Aggregation and Matrix-Wise Freezing (2026.findings-acl)
Copied to clipboard
| Challenge: | Existing methods for fine-tuning large language models often suffer from biased model aggregation and are hindered by significant communication and computation burden. |
| Approach: | They propose a Federated low-rank adaptation system for large language models that leverages pipelined error-mitigated model aggregation and adaptive matrix-wise parameter freezing to mitigate aggregations. |
| Outcome: | The proposed system improves time-to-target by 2.17-8.48 on real-world datasets. |
Knowledge Inheritance for Pre-trained Language Models (2022.naacl-main)
Copied to clipboard
Yujia Qin, Yankai Lin, Jing Yi, Jiajie Zhang, Xu Han, Zhengyan Zhang, Yusheng Su, Zhiyuan Liu, Peng Li, Maosong Sun, Jie Zhou
| Challenge: | Existing large-scale pre-trained language models are mainly trained from scratch individually, ignoring that many well-taught PLMs are available. |
| Approach: | They propose a pre-training framework called knowledge inheritance and propose auxiliary supervision to efficiently learn larger PLMs. |
| Outcome: | The proposed framework can be used to train large-scale language models with huge parameters and a large dataset can be adapted to domain adaptation and knowledge transfer. |
Ranking-Based Autoencoder for Extreme Multi-label Classification (N19-1)
Copied to clipboard
| Challenge: | Existing methods to solve label dependency and noisy labeling problems are limited . experimental results show the proposed method is competitive to state-of-the-art methods . |
| Approach: | They propose a deep learning XML method with word-vector-based self-attention followed by ranking-based AutoEncoder architecture to solve these problems. |
| Outcome: | The proposed method is competitive to state-of-the-art methods on benchmark datasets. |
SPO: Self Preference Optimization with Self Regularization (2025.findings-emnlp)
Copied to clipboard
| Challenge: | Existing reference-free preference optimization methods exhibit higher training efficiency but are prone to overoptimization, leading to performance degradation. |
| Approach: | They propose a reference-free preference optimization method that replaces the logsigmoid loss function with a SiLU function to improve the model's performance. |
| Outcome: | The proposed method achieves 7% improvement over SimPO on AlpacaEval 2 and MT-Bench. |
Tree-Planted Transformers: Unidirectional Transformer Language Models with Implicit Syntactic Supervision (2024.findings-acl)
Copied to clipboard
| Challenge: | Syntactic Language Models (SLMs) have difficulty with inference efficiency due to explicit generation of syntactical structures. |
| Approach: | They propose a method to "plant" trees into attention weights of unidirectional Transformer LMs to implicitly reflect syntactic structures of natural language. |
| Outcome: | The proposed method outperforms SLMs on the SyntaxGym benchmark. |
Safely Learning with Private Data: A Federated Learning Framework for Large Language Model (2024.emnlp-main)
Copied to clipboard
| Challenge: | Existing large language models (LLMs) use large amounts of public data and massive parameters, but private data is often stored in isolated data silos. |
| Approach: | They propose a Federated Learning framework for large language models which offloads most training parameters to the server while training embedding and output layers locally. |
| Outcome: | The proposed framework achieves comparable metrics to centralized chatGLM model on NLU and generation tasks. |
Mixture Content Selection for Diverse Sequence Generation (D19-1)
Copied to clipboard
| Challenge: | Generating diverse sequences exhibit semantically one-to-many relationships between source and target sequences. |
| Approach: | They propose to separate diversification from generation using a general plug-and-play module that wraps around and guides an existing encoder-decoder model. |
| Outcome: | The proposed method shows that diversification and generation are separate steps in the same model and that the model is robust. |
ZeroPrompt: Scaling Prompt-Based Pretraining to 1,000 Tasks Improves Zero-Shot Generalization (2022.findings-emnlp)
Copied to clipboard
| Challenge: | a recent study shows that task scaling can be an efficient alternative to model scaling. |
| Approach: | They propose a multitask pretraining approach ZeroPrompt for zero-shot generalization . they focus on task scaling and zero-shooting to improve model performance . |
| Outcome: | The proposed approach improves zero-shot generalization efficiency by 30 times with task scaling. |
Data Quality Enhancement on the Basis of Diversity with Large Language Models for Text Classification: Uncovered, Difficult, and Noisy (2025.coling-main)
Copied to clipboard
| Challenge: | Existing methods for text classification based on large language models are difficult to apply directly to solve. |
| Approach: | They propose a data quality enhancement method to improve LLMs' performance in classification tasks by using a greedy algorithm to select data and then performing fine-tuning. |
| Outcome: | The proposed method improves the performance of large language models in text classification tasks and significantly improves training efficiency, saving nearly half of the training time. |
CodeTransOcean: A Comprehensive Multilingual Benchmark for Code Translation (2023.findings-emnlp)
Copied to clipboard
| Challenge: | Existing code translation datasets focus on a single pair of programming languages . early software systems are developed using programming languages such as Fortran and COBOL . |
| Approach: | They propose a large-scale comprehensive benchmark that supports the largest variety of programming languages for code translation. |
| Outcome: | The proposed framework supports translations between multiple programming languages and a cross-framework dataset for deep learning code across different frameworks. |
LABO: Towards Learning Optimal Label Regularization via Bi-level Optimization (2023.findings-acl)
Copied to clipboard
| Challenge: | Existing methods for regularizing deep neural networks rely on weight decay, dropout, batch/layer normalization to converge faster and generalize. |
| Approach: | They propose a framework for training with label regularization which includes conventional LS but can also model instance-specific variants. |
| Outcome: | The proposed approach consistently yields better results than conventional regularization on seven machine translation and three image classification tasks while maintaining training efficiency. |
Structural Contrastive Representation Learning for Zero-shot Multi-label Text Classification (2022.findings-emnlp)
Copied to clipboard
| Challenge: | Existing approaches for zero-shot multi-label text classification struggle with accuracy and poor training efficiency. |
| Approach: | They propose a structural contrastive representation learning approach that uses randomized text segmentation to generate high-quality contrastive pairs. |
| Outcome: | The proposed approach improves accuracy and speed up training time on publicly available datasets. |
Learning Better Masking for Better Language Model Pre-training (2023.acl-long)
Copied to clipboard
| Challenge: | Existing PrLMs adopt a Random-Token Masking strategy with a fixed masking ratio and different contents are masked by an equal probability throughout the training. |
| Approach: | They propose two scheduled masking approaches that adaptively tune masking ratio and masked content in different training stages, which improves pre-training efficiency and effectiveness. |
| Outcome: | The proposed methods improve the pre-training efficiency and effectiveness on the downstream tasks. |
BERT Meets CTC: New Formulation of End-to-End Speech Recognition with Pre-trained Masked Language Model (2022.findings-emnlp)
Copied to clipboard
| Challenge: | Existing approaches to connectionist temporal classification (CTC) are based on pre-trained language models (LMs) |
| Approach: | They propose a formulation of connectionist temporal classification that relaxes the conditional independence assumptions used in conventional CTC and incorporates linguistic knowledge through explicit output dependency. |
| Outcome: | The proposed model improves over conventional approaches across variations in speaking styles and languages while maintaining CTC’s training efficiency. |
The Lower The Simpler: Simplifying Hierarchical Recurrent Models (N19-1)
Copied to clipboard
| Challenge: | Using a simplified version of GRU, we replace the GRUs at the middle layers of hierarchical recurrent models with Fixed-size Ordinally-Forgetting Encoding (FOFE). |
| Approach: | They propose to make the lower layers simpler than the upper ones to simplify two typical hierarchical recurrent models, namely Hierarchical Recurrent Encoder-Decoder (HRED) and R-NET, whose basic building block is GRU. |
| Outcome: | The proposed models contain less trainable parameters, consume less training time, and achieve slightly better performance than baseline models. |
Learning from Imperfect Data: Towards Efficient Knowledge Distillation of Autoregressive Language Models for Text-to-SQL (2024.findings-emnlp)
Copied to clipboard
| Challenge: | Existing text-to-SQL LLMs are computationally expensive and difficult to deploy in real-world applications. |
| Approach: | They propose to distill a larger teacher model into a smaller student model by using imperfect data to improve the KD. |
| Outcome: | The proposed method achieves the best tradeoff between performance and efficiency on 5 text-to-SQL benchmarks. |
Feeding What You Need by Understanding What You Learned (2022.acl-long)
Copied to clipboard
| Challenge: | Existing research on machine reading comprehension rely heavily on large-size models and corpus to improve performance. |
| Approach: | They propose a framework that assesses model capabilities in an explainable and multi-dimensional manner. |
| Outcome: | The proposed framework achieves an 11.22% / 8.71% improvement of EM / F1 on MRC tasks. |
DEMix Layers: Disentangling Domains for Modular Language Modeling (2022.naacl-main)
Copied to clipboard
| Challenge: | Extensive experiments with autoregressive transformer LMs show that DEMix layers reduce test-time perplexity and increase training efficiency. |
| Approach: | They introduce a new domain expert mixture layer that enables conditioning a language model on the domain of the input text. |
| Outcome: | Experiments with 1.3B LMs show that DEMix layers reduce test-time perplexity, increase training efficiency, and enable rapid adaptation. |
From Quantity to Quality: Boosting LLM Performance with Self-Guided Data Selection for Instruction Tuning (2024.naacl-long)
Copied to clipboard
Ming Li, Yong Zhang, Zhitao Li, Jiuhai Chen, Lichang Chen, Ning Cheng, Jianzong Wang, Tianyi Zhou, Jing Xiao
| Challenge: | Large Language Models (LLMs) have revolutionized the landscape of artificial intelligence. |
| Approach: | They propose a self-guided method to identify and select cherry samples from open-source datasets, minimizing manual curation and potential cost for instruction tuning an LLM. |
| Outcome: | The proposed method enables LLMs to identify discrepancies between expected responses and intrinsic generation capability, and a marked uptick in model training efficiency. |
Patient Knowledge Distillation for BERT Model Compression (D19-1)
Copied to clipboard
| Challenge: | Pre-trained language models such as BERT have proven to be highly effective for natural language processing tasks, but the high demand for computing resources hinders their application in practice. |
| Approach: | They propose to compress an original large model (teacher) into an equally-effective lightweight shallow network (student) Empirically, this translates into improved results on multiple NLP tasks with a significant gain in training efficiency, without sacrificing model accuracy. |
| Outcome: | The proposed model reduces the computational cost of training models using the teacher model into a lightweight shallow network. |
BPO: Towards Balanced Preference Optimization between Knowledge Breadth and Depth in Alignment (2025.naacl-long)
Copied to clipboard
| Challenge: | Reinforcement Learning with Human Feedback (RLHF) is the key to the success of large language models (LLMs) in recent years. |
| Approach: | They propose a method to balance the number of prompts and responses to improve knowledge breadth and knowledge depth by introducing gradient-based clustering to estimate the knowledge informativeness and usefulness of each augmented sample. |
| Outcome: | The proposed method outperforms baseline methods while maintaining training efficiency. |
RoLoRA: Fine-tuning Rotated Outlier-free LLMs for Effective Weight-Activation Quantization (2024.findings-emnlp)
Copied to clipboard
| Challenge: | Low-Rank Adaptation (LoRA) improves training efficiency by updating only a small portion of the weights in Large Language Models. |
| Approach: | They propose a rotation-aware scheme to fine-tune rotated outlier-free LLMs for effective weight-activation quantization. |
| Outcome: | The proposed method improves low-bit LoRA convergence and post-training quantization robustness. |
Light-PEFT: Lightening Parameter-Efficient Fine-Tuning via Early Pruning (2024.findings-acl)
Copied to clipboard
| Challenge: | Existing methods for fine-tuning large language models are inefficient and redundant . a light-PEFT framework can be used to prune redundant parameters during training . |
| Approach: | They propose a parameter-efficient fine-tuning framework that freezes most parameters of the foundation model and finetuns only a small number of parameters. |
| Outcome: | The proposed framework achieves training and inference speedup, reduces memory usage, and maintains comparable performance and plug-and-play feature of PEFT. |
Dynamic Uncertainty Ranking: Enhancing Retrieval-Augmented In-Context Learning for Long-Tail Knowledge in LLMs (2025.naacl-long)
Copied to clipboard
| Challenge: | Prior work has shown that in-context learning (ICL) with retriever augmentation can help LLMs better capture long-tail knowledge, reducing their reliance on pre-trained data. |
| Approach: | They propose a reinforcement learning-based dynamic uncertainty ranking method that accounts for the varying impact of each retrieved sample on LLM predictions. |
| Outcome: | The proposed method outperforms baseline models on question-answering datasets by 2.76% and 5.96% on long-tail questions that elude zero-shot inference. |
FastCuRL: Curriculum Reinforcement Learning with Stage-wise Context Scaling for Efficient Training R1-like Reasoning Models (2025.findings-emnlp)
Copied to clipboard
| Challenge: | Improving training efficiency remains a challenge in large-scale Reinforcement Learning (RL). |
| Approach: | They propose a curriculum RL framework with stage-wise context scaling to improve RL training efficiency. |
| Outcome: | The proposed framework outperforms state-of-the-art reasoning models on five benchmarks and achieves 49.6% accuracy on AIME 2024. |
Large-Scale Differentially Private BERT (2022.findings-emnlp)
Copied to clipboard
| Challenge: | a recent study shows that scaling up the batch size to millions improves the utility of a DP-SGD step for BERT. |
| Approach: | They propose to use differentially private SGD to pretrain BERT-Large with a batch size of millions to improve the utility of the DP-SGD step. |
| Outcome: | The proposed approach achieves a masked language model accuracy of 60.5% at a batch size of 2M, which is a reasonable privacy setting. |
Federated Learning of Large Language Models with Parameter-Efficient Prompt Tuning and Adaptive Optimization (2023.emnlp-main)
Copied to clipboard
| Challenge: | Prompt tuning of Large Language Models (LLMs) can incur performance degradation or low training efficiency. |
| Approach: | They propose a prompt tuning approach with Adaptive Optimization to enable efficient FL of LLMs. |
| Outcome: | The proposed approach improves performance and efficiency simultaneously and addresses client drift problems on both the device and server sides. |
FPT: Improving Prompt Tuning Efficiency via Progressive Training (2022.findings-emnlp)
Copied to clipboard
| Challenge: | Recent prompt tuning (PT) has gained increasing attention as a parameter-efficient way of tuning pre-trained language models (PLMs). |
| Approach: | They propose a prompt tuning algorithm that uses a small-scale partial PLM and progressively expands its depth and width until the full-model size. |
| Outcome: | The proposed method could save over 30% of training computations while achieving comparable performance. |
CLAPSpeech: Learning Prosody from Text Context with Contrastive Language-Audio Pre-Training (2023.acl-long)
Copied to clipboard
| Challenge: | Existing methods for expressive text-to-speech only implicitly learn prosody with masked token reconstruction tasks. |
| Approach: | They propose a cross-modal contrastive pre-training framework that learns from prosody variance of the same text token under different contexts. |
| Outcome: | The proposed framework can learn from prosody variance of a text token under different contexts. |
Data Selection Curriculum for Abstractive Text Summarization (2023.findings-emnlp)
Copied to clipboard
| Challenge: | Abstractive Text Summarization (ATS) models are commonly trained using large-scale data that is randomly shuffled. |
| Approach: | They propose a data selection curriculum scoring system that measures the learning difficulty of an ATS model and expected performance on an instance. |
| Outcome: | The proposed system surpasses baselines on CNN/DailyMail dataset, utilizing 20% of available instances. |
Improved Unsupervised Chinese Word Segmentation Using Pre-trained Knowledge and Pseudo-labeling Transfer (2023.emnlp-main)
Copied to clipboard
| Challenge: | Existing approaches to unsupervised Chinese word segmentation require multiple inferences to perform word segmenting. |
| Approach: | They propose a method that integrates the segmentation signal from an unsupervised language model to a pre-trained BERT classifier under a pseudo-labeling framework. |
| Outcome: | The proposed method achieves state-of-the-art performance on the eight UCWS tasks while significantly reducing training time compared to previous approaches. |
Quick Back-Translation for Unsupervised Machine Translation (2023.findings-emnlp)
Copied to clipboard
| Challenge: | Unsupervised machine translation models are limited by the run-time of autoregressive inference during back-translation and lack of synthetic data efficiency. |
| Approach: | They propose a two-for-one improvement to Transformer back-translation: Quick Back-Translation (QBT). QBT re-purposes the encoder as a generative model, and uses encoder-generated sequences to train the decoder. |
| Outcome: | Experiments on various WMT benchmarks show that QBT dramatically outperforms standard back-translation only method in terms of training efficiency for comparable translation qualities. |
LongRecipe: Recipe for Efficient Long Context Generalization in Large Language Models (2025.acl-long)
Copied to clipboard
Zhiyuan Hu, Yuliang Liu, Jinman Zhao, Suyuchen Wang, WangYan WangYan, Wei Shen, Qing Gu, Anh Tuan Luu, See-Kiong Ng, Zhiwei Jiang, Bryan Hooi
| Challenge: | Large language models face significant challenges in handling long-context tasks because of their limited effective context window size during pretraining, which restricts their ability to generalize over extended sequences. |
| Approach: | They propose a training strategy for extending the context window of LLMs including impactful token analysis, position index transformation, and training optimization strategies. |
| Outcome: | Experiments on three types of LLMs show that LongRecipe can utilize long sequences while requiring only 30% of the target context window size. |
When Attention Meets Fast Recurrence: Training Language Models with Reduced Compute (2021.emnlp-main)
Copied to clipboard
| Challenge: | Large language models are difficult to train because of the growing computation time and cost. |
| Approach: | They propose a highly-efficient architecture that combines fast recurrence and attention for sequence modeling. |
| Outcome: | The proposed model achieves state-of-the-art on a Wiki-103 and Billion Word datasets using 1.6 days of training on an 8-GPU machine. |
ALDEN: Reinforcement Learning for Active Navigation and Evidence Gathering in Long Documents (2026.acl-long)
Copied to clipboard
| Challenge: | Visually rich documents (VRDs) combine text, tables, and figures within complex, semantically structured layouts. |
| Approach: | They propose a multi-turn reinforcement learning framework that fine-tunes VLMs as interactive agents capable of actively navigating long, visually rich documents. |
| Outcome: | The proposed framework achieves state-of-the-art on five long-document benchmarks. |
Android Coach: Improve Online Agentic Training Efficiency with Single State Multiple Actions (2026.acl-long)
Copied to clipboard
| Challenge: | Existing reinforcement learning methods are expensive due to high latency and sample inefficiency . Currently, RL is limited to one-to-one state-action pairs . |
| Approach: | They propose a framework that shifts the training paradigm to Single State Multiple Actions and introduce a group-wise advantage estimator based on the averaged critic outputs. |
| Outcome: | The proposed framework achieves 7.5% and 8.3% success rate improvements on AndroidLab and AndroidWorld over UI-TARS-1.5-7B and attains 1.4x higher training efficiency than existing methods. |
Outlier-Safe Pre-Training for Robust 4-Bit Quantization of Large Language Models (2025.acl-long)
Copied to clipboard
| Challenge: | Quantization is a practical solution for deploying Large Language Models in resource-constrained environments. |
| Approach: | They propose an outlier-safe pre-training approach that prevents outlier formation . they validate a 1.4B-parameter model on 1 trillion tokens with no outliers . |
| Outcome: | The proposed model achieves a 35.7 average score on 1 trillion tokens with 2% training overhead. |
Mixture-of-Supernets: Improving Weight-Sharing Supernet Training with Architecture-Routed Mixture-of-Experts (2024.findings-acl)
Copied to clipboard
Ganesh Jawahar, Haichuan Yang, Yunyang Xiong, Zechun Liu, Dilin Wang, Fei Sun, Meng Li, Aasish Pappu, Barlas Oguz, Muhammad Abdul-Mageed, Laks Lakshmanan, Raghuraman Krishnamoorthi, Vikas Chandra
| Challenge: | Neural architecture search (NAS) uses weight-sharing supernets to generate diverse subnetworks without retraining. |
| Approach: | They propose a weight-sharing supernet that leverages mixture-of-experts to enhance supernet model expressiveness with minimal training overhead. |
| Outcome: | The proposed method achieves state-of-the-art (SoTA) performance in NAS for fast machine translation models, surpassing NAS-BERT and AutoDistil across various model sizes. |
A Collaborative Multi-agent Reinforcement Learning Framework for Dialog Action Decomposition (2021.emnlp-main)
Copied to clipboard
| Challenge: | Currently, most reinforcement learning methods for dialog policy learning train a centralized agent that selects a predefined joint action concatenating domain name, intent type, and slot name. |
| Approach: | They propose a hierarchical multi-agent framework in which each part of the action is led by a different agent and a joint optimization process that makes agents can exchange their policy information. |
| Outcome: | The proposed framework reduces labor costs for action templates and decreases the size of the action space for each agent. |
OD-RTE: A One-Stage Object Detection Framework for Relational Triple Extraction (2023.acl-long)
Copied to clipboard
| Challenge: | Existing pipelines for relational triple extraction are underutilizing regional information of triple. |
| Approach: | They propose a one-stage Object Detection framework for Relational Triple Extraction . framework uses vertices-based bounding box detection and global relational triple region detection . |
| Outcome: | The proposed framework could extract all types of triples on two widely used datasets. |
Fast Adaptation via Prompted Data: An Efficient Cross-Domain Fine-tuning Method for Large Language Models (2024.lrec-main)
Copied to clipboard
| Challenge: | Large language models (LLMs) have been successful in a variety of natural language understanding tasks, but domain discrepancies between the downstream task and the pre-training corpora may have hindered LLMs to excel further in the vertical applications. |
| Approach: | They propose a Fast Adaptation method for LLMs via Prompted Data that integrates downstream text corpora, gold labels and external knowledge sources into a highly controllable prompt. |
| Outcome: | The proposed method bridges the gap between the downstream task and the pre-training corpora and integrates downstream text corpors, gold labels and external knowledge sources into a highly controllable prompt. |
Let’s Focus on Neuron: Neuron-Level Supervised Fine-tuning for Large Language Model (2025.coling-main)
Copied to clipboard
| Challenge: | Large Language Models (LLMs) are composed of neurons that exhibit diverse behaviors and roles. |
| Approach: | They propose a novel approach that refines the granularity of parameter training down to the individual neuron, enabling a more parameter-efficient fine-tuning model. |
| Outcome: | The proposed approach exceeds the performance of full-parameter fine-tuning and PEFT and provides insights into the analysis of neurons. |
From Complex to Simple: Enhancing Multi-Constraint Complex Instruction Following Ability of Large Language Models (2024.findings-emnlp)
Copied to clipboard
| Challenge: | Large language models (LLMs) follow instructions with elaborate requirements, yet it remains under-explored how to enhance their ability to follow complex instructions with multiple constraints. |
| Approach: | They propose a method to obtain and utilize effective training data to enhance LLMs' ability to follow complex instructions with multiple constraints. |
| Outcome: | The proposed framework improves models' ability to follow instructions generally and generalize effectively across out-of-domain, in domain, and adversarial settings while maintaining general capabilities. |
The Illusion of Specialization: Unveiling the Domain-Invariant "Standing Committee" in Mixture-of-Experts Models (2026.acl-long)
Copied to clipboard
| Challenge: | Mixture of Experts models are widely assumed to achieve domain specialization through sparse routing. |
| Approach: | They propose a framework that analyzes routing behavior at the level of expert groups rather than individual experts. |
| Outcome: | The proposed framework analyzes routing behavior at the level of expert groups rather than individual experts. |
Cross-Modality Relevance for Reasoning on Language and Vision (2020.acl-main)
Copied to clipboard
| Challenge: | Existing approaches to learn and reason over language and vision data for downstream tasks such as visual question answering (VQA) and natural language for visual reasoning (NLVR) |
| Approach: | They propose a cross-modality relevance module that is used in an end-to-end framework to learn the relevance representation between components of various input modalities under supervision of a target task. |
| Outcome: | The proposed approach shows competitive performance on two different language and vision tasks using public benchmarks and improves the state-of-the-art published results. |
Data Augmentation with Hierarchical SQL-to-Question Generation for Cross-domain Text-to-SQL Parsing (2021.emnlp-main)
Copied to clipboard
| Challenge: | Existing approaches to parse text-to-SQL data are lacking labeled data for unseen evaluation databases. |
| Approach: | They propose a framework for enhancing SQL queries by automatically producing large numbers of SQL queries based on an abstract syntax tree grammar. |
| Outcome: | The proposed framework can produce high-quality natural language questions over strong baselines. |
LoRASC: Expressive and Generalizable Low-rank Adaptation for Large Models via Slow Cascaded Learning (2024.findings-emnlp)
Copied to clipboard
| Challenge: | Existing low-rank adaptations have limited expressiveness, a tendency to overfit, and sensitivity to hyperparameter settings. |
| Approach: | They propose a technique to enhance LoRA’s expressiveness and generalization capabilities while preserving its training efficiency. |
| Outcome: | The proposed method outperforms baselines, mitigates overfitting, enhances model stability, and improves OOD robustness. |
Query2Triple: Unified Query Encoding for Answering Diverse Complex Queries over Knowledge Graphs (2023.findings-emnlp)
Copied to clipboard
| Challenge: | Complex Query Answering (CQA) is a challenge task of Knowledge Graphs due to incompleteness of KGs. |
| Approach: | They propose a query embedding approach that decouples the training for simple and complex queries. |
| Outcome: | The proposed approach decouples training for simple and complex queries and achieves state-of-the-art performance over three public benchmarks. |
Exploring Quantization for Efficient Pre-Training of Transformer Language Models (2024.findings-emnlp)
Copied to clipboard
| Challenge: | Quantization has proven to be effective after pre-training and during fine-tuning, but its effects on pre-trainer performance have remained unexplored. |
| Approach: | They propose a linear quantization strategy to be applied during the pre-training of Transformers to improve model efficiency and stability. |
| Outcome: | The proposed method improves model efficiency, stability, and performance while maintaining language modeling ability. |
Federated Data-Efficient Instruction Tuning for Large Language Models (2025.findings-acl)
Copied to clipboard
| Challenge: | Existing federated learning (FL) uses all local data, causing excessive computational overhead and overfitting to local data. |
| Approach: | They propose a federated data-efficient instruction tuning approach which utilizes a representative subset of edge-side data to tune LLMs. |
| Outcome: | The proposed method improves Rouge-L on unseen tasks by 10.72% over the SOTA full-data instruction tuning methods while using less than 1.5% of the data samples. |
Igniting Creative Writing in Small Language Models: LLM-as-a-Judge versus Multi-Agent Refined Rewards (2025.emnlp-main)
Copied to clipboard
| Challenge: | Existing methods for enhancing Large Language Models (LLMs) struggle with novelty and Reinforcement Learning from human feedback (RLHF) is costly. |
| Approach: | They propose to use a Reward Model (RM) and a principle-guided LLM-as-a-Judge to enhance creative output over baselines. |
| Outcome: | The proposed approach significantly enhances creative output over baselines, but the principle-guided LLM-as-a-Judge yields superior generation quality. |
DIVE into MoE: Diversity-Enhanced Reconstruction of Large Language Models from Dense into Mixture-of-Experts (2025.acl-long)
Copied to clipboard
| Challenge: | Existing methods for reconstruction of large language models overlook diversity among experts, leading to potential redundancy. |
| Approach: | They propose a pruning-based expert reconstruction method that prunes a specific LLM and retrains it on routers, experts and normalization modules. |
| Outcome: | The proposed method outperforms pruning and MoE reconstruction methods on Llama-style models with open-source training corpora. |
Memory as Action: Autonomous Context Curation for Long-Horizon Agentic Tasks (2026.findings-acl)
Copied to clipboard
| Challenge: | Existing approaches to managing working memory are based on external mechanisms that lack awareness of the agent’s reasoning state, leading to suboptimal decisions. |
| Approach: | They propose a framework that treats working memory management as learnable policy actions and enables joint optimization of information retention and task performance through end-to-end reinforcement learning. |
| Outcome: | The proposed framework matches models 16 larger while reducing average context length by 51%, with learned strategies that adapt to model capabilities and generalize across task complexities. |
LoopTool: Closing the Data–Training Loop for Robust LLM Tool Calls (2026.acl-long)
Copied to clipboard
| Challenge: | Large Language Models (LLMs) are powerful tools for multi-step tasks, but static data pipelines hinder tool learning and cause noisy labels to persist. |
| Approach: | They propose a fully automated, model-aware data evolution framework that tightly integrates data synthesis and model training. |
| Outcome: | Experiments show that LoopTool-8B significantly surpasses its 32B data generator and achieves new state-of-the-art results on the BFCL-v3 and ACEBench benchmarks for its scale. |
PALM: Few-Shot Prompt Learning for Audio Language Models (2024.emnlp-main)
Copied to clipboard
| Challenge: | Audio-Language Models (ALMs) have recently achieved remarkable success in zero-shot audio recognition tasks, which match features of audio waveforms with class-specific text prompt features. |
| Approach: | They propose a method which optimizes the feature space of the text encoder branch and optimizes audio waveform features with text prompt features. |
| Outcome: | The proposed method outperforms existing methods while being less demanding. |
ImCoref-CeS: An Improved Lightweight Pipeline for Coreference Resolution with LLM-based Checker-Splitter Refinement (2026.acl-long)
Copied to clipboard
Kangyang Luo, Yuzhuo Bai, Shuzheng Si, Cheng Gao, Zhitong Wang, Yingli Shen, Wenhao Li, Zhu Liu, Yufeng Han, Jiayi Wu, Cunliang Kong, Maosong Sun
| Challenge: | Existing supervised neural methods for coreference resolution are underexplored . current methods rely on small language models, but their potential is underexploited . |
| Approach: | They propose a framework that integrates an enhanced supervised model with LLM-based reasoning. |
| Outcome: | The proposed method surpasses existing state-of-the-art methods in coreference resolution. |
One QuantLLM for ALL: Fine-tuning Quantized LLMs Once for Efficient Deployments (2025.acl-long)
Copied to clipboard
| Challenge: | Quantization has shown promise for Large Language Models, but current methods require lengthy training to alleviate quantization loss. |
| Approach: | They propose to decouple weights and incorporate Low-Rank adapters to reduce weight sharing . they validate the approach on LLaMA2 families and Mistral on downstream evaluation . |
| Outcome: | The proposed approach shows high performance while reducing deployment time faced with multiple scenarios. |
AlphaLoRA: Assigning LoRA Experts Based on Layer Training Quality (2024.emnlp-main)
Copied to clipboard
| Challenge: | Recent studies combine LoRA with Mixture-of-Experts (MoE) to improve performance in Large Language Models. |
| Approach: | They propose a method to combine LoRA and Mixture-of-Experts (MoE) to improve performance in Large Language Models. |
| Outcome: | The proposed method reduces redundancy in LoRA experts within the MoE architecture, and improves training quality across layers. |
Scalable Data Ablation Approximations for Language Models through Modular Training and Merging (2024.emnlp-main)
Copied to clipboard
Clara Na, Ian Magnusson, Ananya Harsh Jha, Tom Sherborne, Emma Strubell, Jesse Dodge, Pradeep Dasigi
| Challenge: | Training data compositions for Large Language Models (LLMs) can significantly affect their downstream performance. |
| Approach: | They propose a method which trains individual models on subsets of a training corpus and reuses them across evaluations of combinations of subset. |
| Outcome: | The proposed method improves training efficiency by scaling only linearly with respect to new data. |
Adapters Mixup: Mixing Parameter-Efficient Adapters to Enhance the Adversarial Robustness of Fine-tuned Pre-trained Text Classifiers (2024.emnlp-main)
Copied to clipboard
| Challenge: | Existing studies show that augmenting the training data of pre-trained language models with parametric fine-tuning methods can enhance their robustness under adversarial attacks. |
| Approach: | They propose an approach that fine-tunes PLMs with adapters and adversarial augmentation via mixup to leverage existing knowledge from a set of pre-known attacks. |
| Outcome: | The proposed approach achieves best trade-off between training efficiency and robustness under adversarial attacks compared to baselines on five downstream tasks across six varied black-box attacks and 2 PLMs. |
E3-TIR: Enhanced Experience Exploitation for Tool-Integrated Reasoning (2026.findings-acl)
Copied to clipboard
| Challenge: | Existing training paradigms for Large Language Models (LLMs) suffer from inefficient exploration and mode degradation due to a lack of prior guidance, while SFT-then-RL is limited by high data costs and capability plateaus caused by low-entropy collapse. |
| Approach: | They propose an Enhanced Experience Exploitation paradigm that integrates expert prefixes, expert guided, and self-exploration to improve agent training. |
| Outcome: | The proposed model achieves a 6% performance improvement over traditional paradigms on tool-use tasks while requiring less than 10% of the synthetic data. |
Pre-Training Curriculum for Multi-Token Prediction in Language Models (2025.acl-long)
Copied to clipboard
| Challenge: | Multi-token prediction (MTP) is a pre-training objective for language models . prior work has shown that smaller language models struggle with the MTP objective . |
| Approach: | They propose a curriculum learning strategy that uses multiple prediction heads to predict the next tokens at each prediction step. |
| Outcome: | The proposed curriculum improves performance and output quality while retaining the benefits of self-speculative decoding. |
Demystifying Data Organization for Enhanced LLM Training (2026.acl-long)
Copied to clipboard
Yalun Dai, Yangyu Huang, Tongshen Yang, Yonghan Wang, Xin Zhang, Wenshan Wu, Qihao Zhao, Hao Li, Yuanyuan Gao, Kim-Hui Yap, Scarlett Li
| Challenge: | Large Language Models (LLMs) have revolutionized various fields, yet their training efficiency is heavily reliant on effective data curation. |
| Approach: | They propose to reuse pre-computed sample-level scores originally generated for data efficiency and introduce two new data ordering methods to improve LLM training. |
| Outcome: | The proposed methods improve the stability and performance of LLM training. |
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. |
Mosaic-IT: Cost-Free Compositional Data Synthesis for Instruction Tuning (2025.findings-acl)
Copied to clipboard
| Challenge: | Current instruction tuning relies on teacher models or human intervention to generate and refine the instructions and responses for training, which are costly, non-sustainable, and may lack diversity. |
| Approach: | They propose a human/model-free compositional data synthesis method that can create rich and diverse augmentations from existing instruction tuning data to enhance large language models. |
| Outcome: | The proposed method improves performance over benchmarks and reduces training costs by 80% compared with original instruction tuning. |
MMDEND: Dendrite-Inspired Multi-Branch Multi-Compartment Parallel Spiking Neuron for Sequence Modeling (2025.acl-long)
Copied to clipboard
Kexin Wang, Yuhong Chou, Di Shang, Shijie Mei, Jiahong Zhang, Yanbin Huang, Man Yao, Bo Xu, Guoqi Li
| Challenge: | Vanilla spiking neurons are simplified from complex biological neurons with dendrites, soma, and synapses into single somatic compartments. |
| Approach: | They propose a multi-branch, multi-compartment parallel spiking dendritic neuron with a proportion-adjustable multi-branched structure that enables long-term temporal dependencies. |
| Outcome: | The proposed model achieves better long-sequence modeling capability with fewer parameters and lower energy consumption. |
SDAR-VL: Stable and Efficient Block-wise Diffusion for Vision-Language Understanding (2026.acl-long)
Copied to clipboard
| Challenge: | Existing block-wise discrete diffusion models lack robust autoregressive (AR) decoders. |
| Approach: | They propose a block-wise discrete diffusion framework for large-scale vision-language understanding with a progressive beta noise curriculum. |
| Outcome: | The proposed framework improves training efficiency, convergence stability, and task performance over conventional block diffusion. |
FocusLLM: Precise Understanding of Long Context by Dynamic Condensing (2025.acl-long)
Copied to clipboard
Zhenyu Li, Yike Zhang, Tengyu Pan, Yutao Sun, Zhichao Duan, Junjie Fang, Rong Han, Zixuan Wang, Jianyong Wang
| Challenge: | Existing context condensing methods cannot accurately understand the full context, as there is a considerable amount of information loss in the condensed process. |
| Approach: | They propose a framework to extend the fixed context length of any decoder-only LLM by distilling crucial information from long sequences. |
| Outcome: | The proposed framework extends the fixed context length of any decoder-only LLM, allowing it to focus on relevant information from very long sequences. |
MENTOR: Efficient Autoregressive Image Generation with Balanced Multimodal Control (2026.findings-acl)
Copied to clipboard
| Challenge: | Recent text-to-image models achieve impressive visual quality but still face challenges in precise controllability, balancing multimodal inputs, and high training cost for multimodal image generation. |
| Approach: | They propose an autoregressive framework with a two-stage training paradigm for controllable multimodal image generation. |
| Outcome: | Extensive experiments on DreamBench++ and DreamBech show that the proposed framework achieves a strong balance between textual and visual guidance for controllable image generation. |
TALAS: Teacher-Anchored Layer Alignment with Adaptive Sharpness-Aware Minimization for Embedding Distillation (2026.acl-long)
Copied to clipboard
Quoc Phong Dao, Hoang Son Nguyen, Pham Khanh Chi, Linh Ngo Van, Nguyen Thi Ngoc Diep, Thien Huu Nguyen, Trung Le
| Challenge: | Existing methods that require a student to strictly mimic the teacher’s sentence embeddings or internal features often incur prohibitive computational costs and yield suboptimal performance due to the inherent capacity gap. |
| Approach: | They propose a Teacher-Anchored mechanism that selectively distills final sentence embeddings only into the student’s upper layers, thereby reducing overhead while respecting capacity constraints. |
| Outcome: | Empirical results show that TALAS outperforms existing methods while maintaining high performance. |
LoRACoE: Improving Large Language Model via Composition-based LoRA Expert (2025.emnlp-main)
Copied to clipboard
| Challenge: | Recent studies show that the Mixture of Experts architecture improves performance of large language models. |
| Approach: | They propose a method to build static experts using LoRA parameters . they propose to use rank-level parameters to build experts based on rank-based parameters based in LoRA module. |
| Outcome: | The proposed method improves task performance across a broader range of tasks. |
Flexible-length Text Infilling for Discrete Diffusion Models (2025.emnlp-main)
Copied to clipboard
| Challenge: | Existing discrete diffusion models lack flexibility for text infilling without ground-truth positional data. |
| Approach: | They propose a discrete diffusion model that jointly denoises token values and token positions using a novel sample-level Optimal Transport coupling. |
| Outcome: | The proposed method outperforms existing methods on infilling benchmarks such as One-Billion-Word and Yelp. |
StepHint: Multi-level Stepwise Hints Enhance Reinforcement Learning to Reason (2026.acl-long)
Copied to clipboard
| Challenge: | Reinforcement learning with verifiable rewards (RLVR) approaches face two challenges: the near-miss reward problem and exploration stagnation. |
| Approach: | They propose an algorithm that partitions valid reasoning chains into reasoning steps using multi-level stepwise hints. |
| Outcome: | The proposed method outperforms competing RLVR enhancement methods across six mathematical benchmarks and two out-of-domain benchmarks. |
MTP-RL: Acceleration of Reinforcement Learning Rollouts with Policy-Aligned Multi-Token Prediction (2026.findings-acl)
Copied to clipboard
| Challenge: | Reinforcement learning (RL) is widely applied to boost the performance of pretrained models, yet its training efficiency is severely constrained by rollout generation. |
| Approach: | They propose a framework that accelerates the rollout phase for diverse models by equipping a pipeline to equip the multi-layer parameter-sharing MTP for all models and an advantage-aware MTP optimization strategy. |
| Outcome: | The proposed framework achieves stable growth of acceptance length during RL training, and also accelerates RL rollouts, achieving an average 23.1%–55.3% reduction in rollout time compared to baselines. |
EDSD: Entropy-Driven Design for Faster Speculative Decoding (2026.acl-long)
Copied to clipboard
| Challenge: | Existing methods for speculative decoding incur substantial training overhead to mitigate information misalignment between autoregressive draft model training and decoding. |
| Approach: | They propose an Entropy-Driven Speculative Decoding framework that uses entropy as a unified, interpretable signal for both draft model training and architectural design. |
| Outcome: | Experiments on seven large language models show that EDSD improves training efficiency by 24.8% and increases acceptance length by 4.0% compared to state-of-the-art methods. |
MUTANT: A Recipe for Multilingual Tokenizer Design (2026.acl-long)
Copied to clipboard
| Challenge: | Subword tokenization schemes such as Byte Pair Encoding (BPE) are widely adopted, but their effectiveness in multilingual settings remains understudied. |
| Approach: | They propose a multilingual tokenizer that produces linguistically coherent tokens for multilingual LLMs. |
| Outcome: | The proposed tokenizer improves fertility score by 39.5% over LLaMA4 and 18% over Sutra. |