Papers by Xu Pan

BlindGuard: Safeguarding LLM-based Multi-Agent Systems under Unknown Attacks (2026.acl-long)

Challenge:	Large language models exhibit significant performance discrepancies between high- and low-resource languages.
Approach:	They present an open-source multilingual LLM with 8 billion parameters and a multilingual instruction dataset.
Outcome:	The proposed model achieves consistent multilingual representations across languages.

Copied to clipboard

Rui Miao, Yixin Liu, Yili Wang, Xu Shen, Yue Tan, Yiwei Dai, Shirui Pan, Xin Wang

SafeAgent: Safeguarding LLM Agents via an Automated Risk Simulator (2026.acl-long)

Challenge:	Existing supervised defense methods rely on labeled malicious agents to train a supervised model of malicious behavior.
Approach:	They propose an unsupervised defense method that learns without requiring any attack-specific labels or prior knowledge of malicious behaviors.
Outcome:	The proposed method detects diverse attack types across MAS with various communication patterns while maintaining superior generalizability compared to baselines.

Copied to clipboard

Xueyang Zhou, Weidong Wang, Lin Lu, Jiawen Shi, Guiyao Tie, Xu Yongtian, Lixing Chen, Pan Zhou, Neil Zhenqiang Gong, Lichao Sun

Challenge:	SafeAgent improves agent safety through fully automated synthetic data generation.
Approach:	They propose a framework that improves agent safety through fully automated synthetic data generation.
Outcome:	The proposed framework outperforms closed-source models on two safety benchmarks and one real-world task.

STARS: A Unified Framework for Singing Transcription, Alignment, and Refined Style Annotation (2025.findings-acl)

Copied to clipboard

Wenxiang Guo, Yu Zhang, Changhao Pan, Zhiyuan Zhu, Ruiqi Li, ZheTao Chen, Wenhao Xu, Fei Wu, Zhou Zhao

Speech & Audio in NLP

Challenge:	Existing automated singing annotation (ASA) methods tackle isolated aspects of the annotation pipeline.
Approach:	They propose a framework that addresses transcription, alignment, and refined style annotations.
Outcome:	The proposed framework delivers comprehensive multi-level annotations encompassing: (1) precise phoneme-audio alignment, (2) robust note transcription and temporal localization, (3) expressive vocal technique identification, and (4) global stylistic characterization including emotion and pace.

Aligning VLM Assistants with Personalized Situated Cognition (2025.acl-long)

Copied to clipboard

Yongqi Li, Shen Zhou, Xiaohu Li, Xin Miao, Jintao Wen, Mayi Xu, Jianhao Chen, Birong Pan, Hankun Kang, Yuanyuan Zhu, Ming Zhong, Tieyun Qian

Cool-Fusion: Fuse Large Language Models without Training (2025.acl-long)

Challenge:	Existing studies on vision-language models aligned with general human objectives have not been successful because people with diversified backgrounds have different cognition even in the same situation.
Approach:	They propose to characterize individuals based on the sociological concept of Role-Set and then evaluate their actions to see whether personalized alignment is achieved.
Outcome:	The proposed framework constructs a cognition-aware and action-based reward model for personalized alignment.

Copied to clipboard

Cong Liu, Xiaojun Quan, Yan Pan, Weigang Wu, Xu Chen, Liang Lin

DetGPT: Detect What You Need via Reasoning (2023.emnlp-main)

Challenge:	Cool-Fusion is a simple yet effective approach to combine two or more heterogeneous large language models .
Approach:	They propose a method that fuses the knowledge of two or more heterogeneous large language models to leverage complementary strengths.
Outcome:	The proposed method increases accuracy from three strong source LLMs on GSM8K by 17.4%.

Copied to clipboard

Renjie Pi, Jiahui Gao, Shizhe Diao, Rui Pan, Hanze Dong, Jipeng Zhang, Lewei Yao, Jianhua Han, Hang Xu, Lingpeng Kong, Tong Zhang

Breaking Language Barriers: Cross-Lingual Continual Pre-Training at Scale (2024.emnlp-main)

Challenge:	Recent advances in the field of computer vision have enabled more effective and sophisticated interactions between humans and machines.
Approach:	They propose a reasoning-based object detection paradigm that leverages state-of-the-art multi-modal models and open-vocabulary object detectors to perform reasoning within the context of the user’s instructions and the visual scene.
Outcome:	The proposed method enables users to interact with the system using natural language instructions, allowing for a higher level of interactivity.

Copied to clipboard

Wenzhen Zheng, Wenbo Pan, Xu Xu, Libo Qin, Li Yue, Ming Zhou

AutoMixAlign: Adaptive Data Mixing for Multi-Task Preference Optimization in LLMs (2025.acl-long)

Challenge:	Large Language Models (LLMs) have made significant strides towards Artificial General Intelligence, but training them from scratch is prohibitively expensive.
Approach:	They propose to continuously pre-train LLMs from existing pre-trained LLM models by using a set of parameters instead of randomly initializing them.
Outcome:	The proposed approach saves significant resources and accelerates convergence and performance.

Copied to clipboard

Nicholas E. Corrado, Julian Katz-Samuels, Adithya M Devraj, Hyokun Yun, Chao Zhang, Yi Xu, Yi Pan, Bing Yin, Trishul Chilimbi

MMedAgent: Learning to Use Medical Tools with Multi-modal Agent (2024.findings-emnlp)

Challenge:	Existing approaches to align large language models rely on large ablation studies, heuristics, or human intuition to produce models with strong performance across tasks.
Approach:	They propose an algorithm that mixes datasets during LLM training to balance performance across multiple tasks.
Outcome:	The proposed algorithm outperforms existing methods on multitask alignment setups and achieves convergence rate of O(1/T) in the convex case.

Copied to clipboard

Binxu Li, Tiankai Yan, Yuanting Pan, Jie Luo, Ruiyang Ji, Jiayuan Ding, Zhe Xu, Shilong Liu, Haoyu Dong, Zihao Lin, Yixin Wang

Multimodality

Challenge:	Multi-modal Large Language Models (MLLMs) exhibit limited generality and often fall short when compared to specialized models.
Approach:	They propose a multi-modal medical agent that picks the most suitable medical tools based on user inputs.
Outcome:	The proposed agent performs better than open-source models and the closed-source model, GPT-4o.

HILL: Hierarchy-aware Information Lossless Contrastive Learning for Hierarchical Text Classification (2024.naacl-long)

Copied to clipboard

He Zhu, Junran Wu, Ruomei Liu, Yue Hou, Ze Yuan, Shangzhe Li, Yicheng Pan, Ke Xu

MAIN-RAG: Multi-Agent Filtering Retrieval-Augmented Generation (2025.acl-long)

Challenge:	Existing self-supervised methods in natural language processing rely on augmentation rules to generate contrastive samples.
Approach:	They propose a hierarchy-aware information lossless contrastive learning scheme that uses syntactic information reserved in the input sample and fused during the learning process.
Outcome:	The proposed learning scheme is superior to existing methods in hierarchical text classification . the proposed learning system is based on a structure encoder and a text encoder .

Copied to clipboard

Chia-Yuan Chang, Zhimeng Jiang, Vineeth Rakesh, Menghai Pan, Chin-Chia Michael Yeh, Guanchu Wang, Mingzhi Hu, Zhichao Xu, Yan Zheng, Mahashweta Das, Na Zou

Information Retrieval

Challenge:	Existing RAG systems struggle with the quality of retrieval documents, causing performance degradation and reducing performance.
Approach:	They propose a training-free RAG framework that leverages multiple LLM agents to collaboratively filter and score retrieved documents.
Outcome:	The proposed framework outperforms existing RAG frameworks in QA benchmarks and shows superior answer consistency and answer accuracy over baseline methods.

TheoremQA: A Theorem-driven Question Answering Dataset (2023.emnlp-main)

Copied to clipboard

Wenhu Chen, Ming Yin, Max Ku, Pan Lu, Yixin Wan, Xueguang Ma, Jianyu Xu, Xinyi Wang, Tony Xia

SQL-ASTRA: Alleviating Sparse Feedback in Agentic SQL via Column-Set Matching and Trajectory Aggregation (2026.findings-acl)

Challenge:	Recent LLMs like GPT-4 and PaLM-2 have made tremendous progress in solving fundamental math problems like GSM8K by achieving over 90% accuracy.
Approach:	They propose to use theorem-driven question-answering dataset to evaluate AI models' ability to apply theoretic concepts to solving challenging science problems.
Outcome:	TheoremQA is curated by domain experts and contains 800 high-quality questions covering 350 theoremics from Math, Physics, EE&CS, and Finance.

Copied to clipboard

Long Li, Zhijian Zhou, Jiangxuan Long, Peiyang Liu, Weidi Xu, Zhe Wang, Shirui Pan, Chao Qu

Programming Languages in NLP

Challenge:	Agentic SQL is a framework for multiturn agent learning, but it is limited to single-turn paradigms.
Approach:	They propose a framework that provides a universal two-tiered reward mechanism for credit assignment . they propose 'Aggregated Trajectory Reward' to resolve multi-turn credit assignment.
Outcome:	The proposed framework outperforms SOTA Arctic-Text2SQL-R1-7B on BIRD and Spider 2.0 using identical models.

NOVA: An Iterative Planning Framework for Enhancing Scientific Innovation with Large Language Models (2025.findings-acl)

Copied to clipboard

Xiang Hu, Hongyu Fu, Jinge Wang, Yifeng Wang, Zhikun Li, Renjun Xu, Yu Lu, Yaochu Jin, Lili Pan, Zhenzhong Lan

Alignment for Efficient Tool Calling of Large Language Models (2025.emnlp-main)

Challenge:	Existing approaches to generate research ideas rely on retrieval or prompt engineering to generate ideas.
Approach:	They propose a method that uses iterative planning and search to boost creative potential of LLMs by integrating external knowledge with broader and deeper insights.
Outcome:	The proposed method outperforms the current state-of-the-art in generating 2.5 times more top-rated ideas based on 170 seed papers in a Swiss Tournament evaluation.

Copied to clipboard

Hongshen Xu, Zihan Wang, Zichen Zhu, Lei Pan, Xingyu Chen, Shuai Fan, Lu Chen, Kai Yu

Leveraging Only the Category Name for Aspect Detection through Prompt-based Constrained Clustering (2022.findings-emnlp)

Challenge:	Recent advances in tool learning have enabled large language models to integrate external tools, enhancing their task performance by expanding their knowledge boundaries.
Approach:	They propose a framework that combines probabilistic knowledge boundary estimation with dynamic decision-making to allow LLMs to better assess when to invoke tools based on their confidence.
Outcome:	The proposed framework shows significant improvements in tool efficiency by reducing unnecessary tool usage.

Copied to clipboard

Yazheng Li, Pengyun Wang, Yasheng Wang, Yong Dai, Yadao Wang, Lujia Pan, Zenglin Xu

Low-Resource NLP

Challenge:	Aspect category detection (ACD) aims to automatically identify user-concerned aspects from online reviews.
Approach:	They propose a method that relies on the category name of each aspect and a pretrained language model to generate constraints for clustering.
Outcome:	The proposed framework performs better than existing weakly supervised methods on nine benchmark datasets.

TS-SQL: Test-driven Self-refinement for Text-to-SQL (2025.findings-emnlp)

Copied to clipboard

Wenbo Xu, Haifeng Zhu, Liang Yan, Chuanyi Liu, Peiyi Han, Shaoming Duan, Jeff Z. Pan

Programming Languages in NLP

Challenge:	null
Approach:	null
Outcome:	null

Prototype Tuning: A Meta-Learning Approach for Few-Shot Document-Level Relation Extraction with Large Language Models (2025.findings-naacl)

Copied to clipboard

Dinghao Pan, Yuanyuan Sun, Bo Xu, Jiru Li, Zhihao Yang, Ling Luo, Hongfei Lin, Jian Wang

PsychEval: A Multi-Session and Multi-Therapy Benchmark for High-Realism AI Psychological Counselor (2026.findings-acl)

Challenge:	Few-Shot Document-Level Relation Extraction (FSDLRE) aims to develop models capable of generalizing to new categories with minimal support examples.
Approach:	They propose a meta-training approach to train Large Language Models to improve their ICL capabilities . they construct simulated episodes using relation types that do not overlap with test corpus .
Outcome:	Experimental results show that the proposed approach outperforms baseline models on few-shot tasks.

Copied to clipboard

Qianjun Pan, Junyi Wang, Jie Zhou, Yutao Yang, Junsong Li, Kaiyin Xu, Yougen Zhou, Yihan Li, JingYuan Zhao, Qin Chen, Ningning Zhou, Kai Chen, Liang He

Challenge:	Existing models focus on a single therapy, but complex cases require flexible strategies among various therapies.
Approach:	They propose a multi-session, multi-therapy, and highly realistic benchmark . it is designed to address three key challenges: 1) can we train a highly realistic AI counselor? 2) How to systematically evaluate an AI counselor?"
Outcome:	The proposed benchmark is annotated with extensive professional skills and includes over 677 meta-skills and 4577 atomic skills.

Aristotle: Mastering Logical Reasoning with A Logic-Complete Decompose-Search-Resolve Framework (2025.acl-long)

Copied to clipboard

Jundong Xu, Hao Fei, Meng Luo, Qian Liu, Liangming Pan, William Yang Wang, Preslav Nakov, Mong-Li Lee, Wynne Hsu

Responsible & Trustworthy NLP

Challenge:	Existing systems fail to fully leverage the structure of logical tasks throughout the reasoning process, causing bottlenecks in efficiency and efficacy.
Approach:	They propose a logic-complete reasoning framework, Aristotle, which integrates symbolic expressions and logical rules into the entire reasoning process.
Outcome:	The proposed framework outperforms state-of-the-art reasoning frameworks in accuracy and efficiency.

User-Assistant Bias in LLMs (2026.findings-acl)

Copied to clipboard

Xu Pan, Jingxuan Fan, Zidi Xiong, Ely Hahami, Jorin Overwiening, Ziqian Xie

A Survey on Training-free Alignment of Large Language Models (2025.findings-emnlp)

Challenge:	Modern large language models are typically trained using structured role tags . asymmetries in training data associated with different role tags can potentially introduce inductive biases.
Approach:	They propose a task-agnostic benchmark to test user–assistant bias in large language models . they find human-preference alignment amplifies user bias, reasoning fine-tuning reduces it .
Outcome:	The proposed benchmark tests show that most instruction-tuned models exhibit strong user bias . human-preference alignment amplifies user bias, while reasoning fine-tuning reduces it.

Copied to clipboard

Birong Pan, Yongqi Li, Weiyu Zhang, Wenpeng Lu, Mayi Xu, Shen Zhou, Yuanyuan Zhu, Ming Zhong, Tieyun Qian

LLM-Based Multi-Agent Systems are Scalable Graph Generative Models (2025.findings-acl)

Challenge:	a survey of large language models (LLMs) aims to ensure outputs adhere to human values, ethical standards, and legal norms.
Approach:	They present the first systematic review of TF alignment methods . they categorize them by stages of pre-decoding, in-decoder and post-decoration .
Outcome:	The proposed methods are based on training-free (TF) alignment techniques . they are able to be used in open-source and closed-source environments without retraining .

Copied to clipboard

Jiarui Ji, Runlin Lei, Jialing Bi, Zhewei Wei, Xu Chen, Yankai Lin, Xuchen Pan, Yaliang Li, Bolin Ding

Multimodality

Challenge:	Social graphs are mathematical structures stem from pairwise interactions between entities through nodes and edges.
Approach:	They propose a framework for dynamic, text-attributed social graph generation that simulates the temporal node and edge generation processes for zero-shot social graphs.
Outcome:	The proposed framework improves macroscopic graph structure metrics by 11% . the proposed model can generate graphs with up to 100,000 nodes or 10 million edges .

Multi-Instance Multi-Label Learning Networks for Aspect-Category Sentiment Analysis (2020.emnlp-main)

Copied to clipboard

Yuncong Li, Cunxiang Yin, Sheng-hua Zhong, Xu Pan

Sentiment Analysis

Challenge:	Existing methods to detect sentiment toward aspect categories ignore the fact that the sentiment of an aspect category mentioned in a sentence is an aggregation of the sentiments of the words indicating the aspect category in the sentence, which leads to suboptimal performance.
Approach:	They propose a multi-instance multi-label learning network for Aspect-Category sentiment analysis that treats sentences as bags, words as instances, and the words indicating an aspect category as key instances of the aspect category.
Outcome:	The proposed model is based on three public datasets showing that it performs well.

Fire Burns, Sword Cuts: Commonsense Inductive Bias for Exploration in Text-based Games (2022.acl-short)

Copied to clipboard

Dongwon Ryu, Ehsan Shareghi, Meng Fang, Yunqiu Xu, Shirui Pan, Reza Haf

GroundingGPT: Language Enhanced Multi-modal Grounding Model (2024.acl-long)

Challenge:	Existing RL agents are far away from solving text-based games due to their combinatorially large action spaces that hinders efficient exploration.
Approach:	They propose an exploration technique that injects external commonsense knowledge, via a pretrained language model, into the agent during training when the agent is the most uncertain about its next action.
Outcome:	The proposed method exhibits improvement on the collected game scores during the training in four out of nine games from Jericho.

Copied to clipboard

Zhaowei Li, Qi Xu, Dong Zhang, Hang Song, YiQing Cai, Qi Qi, Ran Zhou, Junting Pan, Zefeng Li, Vu Tu, Zhida Huang, Tao Wang

INSTRUCTSCORE: Towards Explainable Text Generation Evaluation with Automatic Feedback (2023.emnlp-main)

Challenge:	Existing multi-modal large language models focus on capturing global information while neglecting the fine-grained local information in multimodal inputs.
Approach:	They propose an end-to-end language enhanced multi-modal grounding model that performs fine-grained grounding tasks for image, video and audio.
Outcome:	The proposed model achieves impressive fine-grained understanding of multi-modal inputs while maintaining or improving its global comprehension capabilities.

Copied to clipboard

Wenda Xu, Danqing Wang, Liangming Pan, Zhenqiao Song, Markus Freitag, William Wang, Lei Li

CMMaTH: A Chinese Multi-modal Math Skill Evaluation Benchmark for Foundation Models (2025.coling-main)

Challenge:	Existing methods to evaluate the quality of language generation do not provide explicit explanation of their verdicts.
Approach:	They propose a fine-grained explainable evaluation metric for text generation that harnesses human instruction and implicit knowledge of GPT-4 to fine-tune it.
Outcome:	The proposed model outperforms all other unsupervised metrics on translation, captioning, data-to-text, and commonsense generation tasks.

Copied to clipboard

Zhongzhi Li, Ming-Liang Zhang, Pei-Jie Wang, Jian Xu, Rui-Song Zhang, Yin Fei, Zhi-Long Ji, Jin-Feng Bai, Zhen-Ru Pan, Jiaxin Zhang, Cheng-Lin Liu

Numerical Reasoning

Challenge:	Large language models excel in various language tasks, while large multimodal models effectively handle visual-language problems.
Approach:	They propose to use a multimodal multimodal model evaluation benchmark to evaluate model performance in Chinese K12 classrooms.
Outcome:	The proposed model evaluation tool is integrated with the CMMaTH dataset.

FR-Spec: Accelerating Large-Vocabulary Language Models via Frequency-Ranked Speculative Sampling (2025.acl-long)

Copied to clipboard

Weilin Zhao, Tengyu Pan, Xu Han, Yudi Zhang, Sun Ao, Yuxiang Huang, Kaihuo Zhang, Weilun Zhao, Yuxuan Li, Jie Zhou, Hao Zhou, Jianyong Wang, Maosong Sun, Zhiyuan Liu

Don’t Act Blindly: Robust GUI Automation via Action-Effect Verification and Self-Correction (2026.acl-long)

Challenge:	Speculative sampling is an efficient way to accelerate the auto-regressive generation process of large language models.
Approach:	They propose a frequency-ranked speculative sampling framework that optimizes draft candidate selection through vocabulary space compression.
Outcome:	Experiments show that FR-Spec reduces LM Head computation overhead by 75% while ensuring the equivalence of the final output distribution.

Copied to clipboard

Yuzhe Zhang, Xianwei Xue, Xingyong Wu, Mengke Chen, Chen Liu, Xinran He, Run Shao, Feiran Liu, Huanmin Xu, Qiutong Pan, Haiwei Wang

pFedGPT: Hierarchically Optimizing LoRA Aggregation Weights for Personalized Federated GPT Models (2025.emnlp-main)

Challenge:	Existing GUI agents assume deterministic environment responses, generating actions without verifying whether previous operations succeeded.
Approach:	They propose a GUI agent that explicitly models action outcomes and recovery under noisy environments.
Outcome:	The proposed agent reduces failure loops and improves recovery success in noisy environments while maintaining competitive standard task performance.

Copied to clipboard

Zhanming Shen, Tianqi Xu, Hao Wang, Jian Li, Miao Pan

Distinguish Confusing Law Articles for Legal Judgment Prediction (2020.acl-main)

Challenge:	Existing methods for fine-tuning Large Language Models (LLMs) struggle with data heterogeneity and adapt shared global knowledge to individual client needs.
Approach:	They propose a framework that leverages Hierarchical Bayesian Optimization (HBO) for fine-grained, personalized LoRA aggregation.
Outcome:	The proposed framework achieves state-of-the-art (SOTA) performance on personalized FL benchmarks while introducing only minimal (approx. 4%) additional optimization overhead.

Copied to clipboard

Nuo Xu, Pinghui Wang, Long Chen, Li Pan, Xiaoyan Wang, Junzhou Zhao

Information Extraction & Text Mining

Challenge:	Existing methods to assist legal judgment are limited and can't solve confusing charges issue.
Approach:	They propose an end-to-end model to predict a legal judgment based on a textual description of the case and a graph neural network to learn subtle differences between confusing law articles.
Outcome:	The proposed model can learn subtle differences between confusing law articles and extract effective discriminative features from fact descriptions.

DMSD: Dual-Modal Semantic Disentanglement for Compositional Zero-Shot Learning (2026.findings-acl)

Copied to clipboard

Pan Yang, Jing Yang, Ruan Xiao li, Yuling Chen, Yuankai Wu, Quan Zhou, Xu Wang

Illusions of Confidence? Diagnosing LLM Truthfulness via Neighborhood Consistency (2026.acl-long)

Challenge:	Compositional Zero-Shot Learning (CZSL) is a new research paradigm that learns sub-concepts from seen compositions and recognizes unseen novel combinations.
Approach:	They propose a Dual-Modal Semantic Disentanglement framework that integrates visual and textual information to achieve effective sub-concept disentangling.
Outcome:	The proposed framework achieves state-of-the-art performance on three benchmark datasets . it integrates a class-centroid bridge module to guide class centroids toward the textual space .

Copied to clipboard

Haoming Xu, Ningyuan Zhao, Yunzhi Yao, Weihong Xu, Hongru Wang, Xinle Deng, Shumin Deng, Jeff Z. Pan, Huajun Chen, Ningyu Zhang

Challenge:	Existing evaluations rely on point-wise confidence, which can mask brittle belief.
Approach:	They propose a measure of belief robustness that evaluates coherence across a conceptual neighborhood.
Outcome:	The proposed model is more resistant to interference than existing models.

Good Reasoning Makes Good Demonstrations: Implicit Reasoning Quality Supervision via In-Context Reinforcement Learning (2026.findings-acl)

Copied to clipboard

Tiehua Mei, Minxuan Lv, Leiyu Pan, Zhenpeng Su, Hongru Hou, Hengrui Chen, Ao Xu, Deqing Yang

Reasoning

Challenge:	Reinforcement Learning with Verifiable Rewards (RLVR) improves reasoning in large language models but treats all correct solutions equally, potentially reinforcing flawed traces that arrive at correct answers by chance.
Approach:	They propose a method that reweights rewards by a factor approximately proportional to Evidence Gain and assigns higher weights to high-quality traces without requiring costly computation.
Outcome:	Experiments on mathematical reasoning benchmarks show that Reinforcement Learning with Verifiable Rewards (RLVR) improves reasoning in large language models but treats all correct solutions equally.

An ensemble CNN method for biomedical entity normalization (D19-57)

Copied to clipboard

Pan Deng, Haipeng Chen, Mengyao Huang, Xiaowen Ruan, Liang Xu

Knowledge Representation

Challenge:	Named entity recognition (NER) and entity normalization (entity linking) are two fundamental natural language processing tasks to achieve entity normalizing.
Approach:	They propose a CNN method that normalizes microbiology-related entities to concepts in standard dictionaries.
Outcome:	The proposed method performs well in the BioNLP-OST19 shared task Bacteria Biotope.

OpenResearcher: Unleashing AI for Accelerated Scientific Research (2024.emnlp-demo)

Copied to clipboard

Yuxiang Zheng, Shichao Sun, Lin Qiu, Dongyu Ru, Cheng Jiayang, Xuefeng Li, Jifan Lin, Binjie Wang, Yun Luo, Renjie Pan, Yang Xu, Qingkai Min, Zizhao Zhang, Yiwen Wang, Wenjie Li, Pengfei Liu

Information Retrieval

Challenge:	Global scientific publications are growing annually by about 4%-5% (Pinedo et al., 2024).
Approach:	They introduce an AI-assisted platform that answers diverse questions from researchers using Retrieval-Augmented Generation (RAG) they develop various tools to understand queries, search from the scientific literature, filter retrieved information, provide accurate and comprehensive answers, and self-refine answers.
Outcome:	OpenResearcher is built on Retrieval-Augmented Generation (RAG) to integrate Large Language Models (LLMs) with up-to-date, domain-specific knowledge.

Merge Hijacking: Backdoor Attacks to Model Merging of Large Language Models (2025.acl-long)

Copied to clipboard

Zenghui Yuan, Yangming Xu, Jiawen Shi, Pan Zhou, Lichao Sun

ScaleBiO: Scalable Bilevel Optimization for LLM Data Reweighting (2025.acl-long)

Challenge:	Existing research on model merging focuses on optimizing model performance and minimizing backdoors.
Approach:	They propose a backdoor attack targeting model merging in Large Language Models that creates a unified model for multi-domain tasks.
Outcome:	The proposed attack is effective across models, merging algorithms, and tasks while maintaining utility across tasks.

Copied to clipboard

Rui Pan, Dylan Zhang, Hanning Zhang, Xingyuan Pan, Minrui Xu, Jipeng Zhang, Renjie Pi, Xiaoyu Wang, Tong Zhang

Green & Sustainable NLP

Challenge:	Existing paradigms for bilevel optimization require second-order information, making it difficult to scale them up.
Approach:	They propose a scalable instantiation of a bilevel optimization paradigm for large-scale LLMs by using a memory-efficient training technique.
Outcome:	The proposed paradigm scales to 30B-sized LLMs on 8H100 GPUs.

Model Editing Harms General Abilities of Large Language Models: Regularization to the Rescue (2024.emnlp-main)

Copied to clipboard

Jia-Chen Gu, Hao-Xiang Xu, Jun-Yu Ma, Pan Lu, Zhen-Hua Ling, Kai-Wei Chang, Nanyun Peng

ALinFiK: Learning to Approximate Linearized Future Influence Kernel for Scalable Third-Parity LLM Data Valuation (2025.naacl-long)

Challenge:	Existing methods that edit large language models with updated knowledge can cause side effects on the general abilities of LLMs such as reasoning, natural language inference, and question answering.
Approach:	They propose to regularize the edit update weights by imposing constraints on their complexity based on the RElative Change in weighT.
Outcome:	The proposed method can significantly mitigate the side effects while maintaining over 94% editing performance.

Copied to clipboard

Yanzhou Pan, Huawei Lin, Yide Ran, Jiamin Chen, Xiaodong Yu, Weijie Zhao, Denghui Zhang, Zhaozhuo Xu

Rethinking the Video Sampling and Reasoning Strategies for Temporal Sentence Grounding (2022.findings-emnlp)

Challenge:	Large Language Models (LLMs) heavily rely on high-quality training data, making data valuation crucial for optimizing model performance.
Approach:	They propose a third-party data valuation approach that assesses the value of individual data samples and proposes a learning strategy to approximate LinFiK.
Outcome:	The proposed approach surpasses baselines in effectiveness and efficiency, showing significant scalability advantages as LLM parameters increase.

Copied to clipboard

Jiahao Zhu, Daizong Liu, Pan Zhou, Xing Di, Yu Cheng, Song Yang, Wenzheng Xu, Zichuan Xu, Yao Wan, Lichao Sun, Zeyu Xiong

Multimodality

Challenge:	Existing methods for temporal sentence grounding ignore two crucial issues . 1) Boundary-bias: the video downsampling process may lose these two frames . 2) Reasoning-biases: such incorrect new boundary frames lead to the reasoning bias .
Approach:	They propose a siamese sampling mechanism to generate additional contextual frames . they use a reasoning strategy to learn the inter-relationship among these frames a .
Outcome:	Extensive experiments demonstrate the effectiveness of a new siamese sampling network on three challenging datasets.

Topology-Enhanced Alignment for Large Language Models: Trajectory Topology Loss and Topological Preference Optimization (2026.findings-acl)

Copied to clipboard

Yurui Pan, Ke Xu, Bo Peng

SciAgent: Tool-augmented Language Models for Scientific Reasoning (2024.emnlp-main)

Challenge:	Existing algorithms for supervised fine-tuning and reinforcement learning from human feedback (RLHF) do not constrain how hidden states move from a user prompt to an answer.
Approach:	They propose a topology-enhanced alignment framework that regularizes these trajectories using 0-dimensional persistent homology.
Outcome:	The proposed framework regularizes semantic trajectory in hidden space using 0-dimensional persistent homology.

Copied to clipboard

Yubo Ma, Zhibin Gou, Junheng Hao, Ruochen Xu, Shuohang Wang, Liangming Pan, Yujiu Yang, Yixin Cao, Aixin Sun

Beyond Itinerary Planning—A Real-World Benchmark for Multi-Turn and Tool-Using Travel Tasks (2026.acl-long)

Challenge:	SciAgent surpasses other LLMs with the comparable size by more than 8.0% in absolute accuracy.
Approach:	They propose a tool-augmented scientific reasoning setting that supplements LLMs with scalable toolsets and builds a benchmark to evaluate LLM’s abilities with tool assistance.
Outcome:	The proposed setting augments LLMs with scalable toolsets and shifts the focus from pursuing an omniscient problem solver to a proficient tool-user.

Copied to clipboard

Xiang Cheng, Yulan Hu, Xiangwen Zhang, Lu Xu, Lide Tan, Zheng Pan, Xin Li, Yong Liu

Responsible & Trustworthy NLP

Challenge:	Existing studies on LLM performance on travel planning have shown that existing settings are limited due to limited domain coverage, insufficient modeling of users’ implicit preferences in multi-turn conversations, and a lack of evaluation of agents’ capability boundaries.
Approach:	They propose a benchmark to evaluate LLMs' planning and tool-use abilities in real-world settings by collecting user queries, user preferences, and tools from real scenarios.
Outcome:	The proposed benchmark evaluates agents' capabilities in real-world settings and shows that even advanced models exhibit imbalanced performance across different capabilities.

Faithful Logical Reasoning via Symbolic Chain-of-Thought (2024.acl-long)

Copied to clipboard

Jundong Xu, Hao Fei, Liangming Pan, Qian Liu, Mong-Li Lee, Wynne Hsu

Machine Translation

Challenge:	SymbCoT is a framework that integrates symbolic expressions and logic rules with CoT prompting.
Approach:	They propose a Symbolic Chain-of-Thought framework that integrates symbolic expressions and logic rules with CoT prompting.
Outcome:	The proposed framework improves on 5 standard datasets with symbolic expressions and rules . it shows that it is more faithful, flexible, and explainable than the current method .

Rescorla-Wagner Steering of LLMs for Undesired Behaviors over Disproportionate Inappropriate Context (2025.emnlp-main)

Copied to clipboard

Rushi Wang, Jiateng Liu, Cheng Qian, Yifan Shen, Yanzhou Pan, Zhaozhuo Xu, Ahmed Abbasi, Heng Ji, Denghui Zhang

Reasoning Fails Where Step Flow Breaks (2026.acl-long)

Challenge:	Incorporating external context can enhance the response quality of Large Language Models (LLMs). however, real-world contexts often mix relevant information with disproportionate inappropriate content.
Approach:	They propose a Poisoned Context Testbed to pair queries with real-world contexts . they propose 'rw-Steering' to internalize inappropriate signals .
Outcome:	The proposed model improves response quality by 39.8% and reverses undesirable behavior curve.

Copied to clipboard

Xiaoyu Xu, Yulan Pan, Xiaosong Yuan, Zhihong Shen, Minghao Su, Yuanhao Su, Xiaofeng Zhang

Reasoning

Challenge:	Existing analysis tools struggle with long chain of thought traces.
Approach:	They propose a saliency-inspired test-time intervention that adjusts shallow saliencies to improve accuracy on math, science, and coding tasks.
Outcome:	The proposed model improves accuracy on math, science, and coding tasks without retraining.

GenTool: Enhancing Tool Generalization in Language Models through Zero-to-One and Weak-to-Strong Simulation (2025.findings-acl)

Copied to clipboard

Jie He, Jennifer Neville, Mengting Wan, Longqi Yang, Hui Liu, Xiaofeng Xu, Xia Song, Jeff Z. Pan, Pei Zhou