Papers by Hui Shen
Red-Teaming LLM Multi-Agent Systems via Communication Attacks (2025.findings-acl)
Copied to clipboard
| Challenge: | Large Language Model-based Multi-Agent Systems (LLM-MAS) have revolutionized complex problem-solving capability by enabling agent collaboration through message-based communications. |
| Approach: | They propose an attack that exploits communication mechanisms in Large Language Model-based Multi-Agent Systems (LLM-MAS) by intercepting and manipulating inter-agent messages. |
| Outcome: | The proposed attack exploits communication mechanisms in large language model-based multi-agent systems by intercepting and manipulating inter-agencies. |
LLM as Effective Streaming Processor: Bridging Streaming-Batch Mismatches with Group Position Encoding (2025.findings-acl)
Copied to clipboard
| Challenge: | Existing methods for adapting LLMs to streaming rely on expensive re-encoding or limited scalability. |
| Approach: | They propose a group position encoding paradigm built on batch architectures to enhance consistency between streaming and batch modes. |
| Outcome: | The proposed method outperforms existing methods on cross-lingual and cross-modal tasks. |
Unveiling the Key Factors for Distilling Chain-of-Thought Reasoning (2025.findings-acl)
Copied to clipboard
Xinghao Chen, Zhijing Sun, Guo Wenjin, Miaoran Zhang, Yanjun Chen, Yirong Sun, Hui Su, Yijie Pan, Dietrich Klakow, Wenjie Li, Xiaoyu Shen
| Challenge: | Large Language Models (LLMs) excel in reasoning tasks through Chain-of-Thought prompting. |
| Approach: | They examine the factors influencing CoT distillation including granularity, format and teacher model. |
| Outcome: | The proposed model is based on four teacher models and seven student models across seven mathematical and commonsense reasoning datasets. |
Locate, Steer, and Improve: A Practical Survey of Actionable Mechanistic Interpretability in Large Language Models (2026.findings-acl)
Copied to clipboard
Hengyuan Zhang, Zhihao Zhang, Ercong Nie, Mingyang Wang, Zunhai Su, Yiwei Wang, Qianli Wang, Shuzhou Yuan, Xufeng Duan, Qibo Xue, Zeping Yu, Chenming Shang, Xiao Liang, Jing Xiong, Hui Shen, Chaofan Tao, Zhengwu Liu, Senjie Jin, Zhiheng Xi, Dongdong Zhang, Sophia Ananiadou, Tao Gui, Ruobing Xie, Hayden Kwok-Hay So, Hinrich Schuetze, Xuanjing Huang, Qi Zhang, Ngai Wong
| Challenge: | Existing literature on mechanistic interpretation (MI) treats it as an observational science, leaving practical applications underexplored. |
| Approach: | They propose a survey structured around the pipeline to identify and improve MI models. |
| Outcome: | The proposed framework enables tangible improvements in Alignment, Capability, and Efficiency. |
NEXUS Network: Connecting the Preceding and the Following in Dialogue Generation (D18-1)
Copied to clipboard
| Challenge: | Sequence-to-Sequence models favor short generic responses . however, the model is not suitable for modeling dialogues . |
| Approach: | They propose a model that connects preceding and following conversations to a prior distribution to avoid non-differentiability of discrete natural language tokens. |
| Outcome: | The proposed model is highly efficient in learning the backbone of human-computer communications, but favors short generic responses. |
Improving Multi-turn Dialogue Modelling with Utterance ReWriter (P19-1)
Copied to clipboard
| Challenge: | Recent research has achieved impressive results in single-turn dialogue modelling, but multi-turn models still remain challenging. |
| Approach: | They propose to rewrite human utterances as a pre-process to help multi-turn dialgoue modelling. |
| Outcome: | The proposed architecture achieves remarkably good performance on the utterance rewriting task. |
VisiPruner: Decoding Discontinuous Cross-Modal Dynamics for Efficient Multimodal LLMs (2025.emnlp-main)
Copied to clipboard
| Challenge: | Multimodal Large Language Models (MLLMs) suffer from significant computational overhead due to the quadratic growth of attention computations with the number of multimodal tokens. |
| Approach: | They propose a training-free pruning framework that prunes multimodal tokens without a trained pruning method. |
| Outcome: | The proposed pruning framework outperforms existing token pruning methods and generalizes across diverse MLLMs. |
LDIR: Low-Dimensional Dense and Interpretable Text Embeddings with Relative Representations (2025.findings-acl)
Copied to clipboard
| Challenge: | Existing text embeddings with high dimensions are difficult to trace and interpret. |
| Approach: | They propose low-dimensional and interpretable text embeddings with relative representations that encode semantic meanings in a vector space where similar texts are close together in the representation space. |
| Outcome: | The proposed embeddings outperform existing models on multiple tasks with fewer dimensions and are lowdimensional and dense while maintaining interpretability. |
Neural Data-to-Text Generation via Jointly Learning the Segmentation and Correspondence (2020.acl-main)
Copied to clipboard
| Challenge: | Recent neural attention models conflate all steps into a single end-to-end system and simplify training process. |
| Approach: | They propose to explicitly segment target text into fragment units and align them with their data correspondences. |
| Outcome: | The proposed model outperforms neural attention models on E2E and WebNLG benchmarks. |
ErrorRadar: Benchmarking Complex Mathematical Reasoning of Multimodal Large Language Models Via Error Detection (2026.findings-acl)
Copied to clipboard
Yibo Yan, Shen Wang, Jiahao Huo, Hang Li, Boyan Li, Jiamin Su, Xiong Gao, YiFan Zhang, Tianlong Xu, Zhendong Chu, Aoxiao Zhong, Kun Wang, Hui Xiong, Philip S. Yu, Xuming Hu, Qingsong Wen
| Challenge: | Current mathematical benchmarks focus on evaluating MLLMs’ problem-solving ability, yet there is a crucial gap in addressing more complex scenarios such as error detection. |
| Approach: | They propose to evaluate multimodal error detection by evaluating two sub-tasks error step identification and error categorization. |
| Outcome: | The proposed task evaluates MLLMs' ability to handle multimodal questions compared to text-only models. |
Seeing but Not Thinking: Routing Distraction in Multimodal Mixture-of-Experts (2026.acl-long)
Copied to clipboard
Haolei Xu, Haiwen Hong, Hongxing Li, Rui Zhou, Yang Zhang, Longtao Huang, Hui Xue, Yongliang Shen, Weiming Lu, Yueting Zhuang
| Challenge: | Existing multimodal Mixture-of-Experts models accurately perceive image content yet fail in subsequent reasoning . Seeing but not thinking phenomenon is a puzzling phenomenon . |
| Approach: | They propose a routing-guided intervention method that enhances domain expert activation. |
| Outcome: | The proposed method achieves consistent improvements on visual reasoning tasks. |
Select and Attend: Towards Controllable Content Selection in Text Generation (D19-1)
Copied to clipboard
| Challenge: | Recent neural network models conflate content selection and surface realization into a black-box architecture, resulting in content to be described in text cannot be explicitly controlled. |
| Approach: | They propose to decouple content selection from the decoder to allow finer-grained control over the generation. |
| Outcome: | The proposed model can be trained end-to-end without human annotations and achieves promising results in data-totext and headline generation tasks. |
MovieChats: Chat like Humans in a Closed Domain (2020.emnlp-main)
Copied to clipboard
| Challenge: | Currently, open-domain chatbots are far from satisfactory. |
| Approach: | They propose a unified, readily scalable neural approach which reconciles all subtasks like intent prediction and knowledge retrieval. |
| Outcome: | The proposed approach outperforms commercial systems replying on complex rules on static and interactive tests and shows that the results are remarkably good. |
MultiConIR: Towards Multi-Condition Information Retrieval (2025.findings-emnlp)
Copied to clipboard
Xuan Lu, Sifan Liu, Bochao Yin, Yongqi Li, Xinghao Chen, Hui Su, Yaohui Jin, Wenjun Zeng, Xiaoyu Shen
| Challenge: | MultiConIR is a benchmark designed to evaluate retrieval and reranking models under nuanced multi-condition query scenarios. |
| Approach: | They propose a benchmark to evaluate retrieval and reranking models under nuanced multi-condition query scenarios. |
| Outcome: | The proposed benchmark evaluates retrieval and reranking models under nuanced multi-condition query scenarios across five domains. |
Improving Latent Alignment in Text Summarization by Generalizing the Pointer Generator (D19-1)
Copied to clipboard
| Challenge: | Modern pointer generators only capture exact word matches, ignoring possible inflections or abstractions, which restricts its power of capturing richer latent alignment. |
| Approach: | They propose a pointer generator architecture that allows the model to "edit" pointed tokens instead of always copying them. |
| Outcome: | The proposed model captures more latent alignment relations than exact word matches and generates higher-quality summaries validated by both qualitative and quantitative evaluations. |
RoCBert: Robust Chinese Bert with Multimodal Contrastive Pretraining (2022.acl-long)
Copied to clipboard
| Challenge: | Large-scale pretrained language models have achieved SOTA results on NLP tasks but are vulnerable to adversarial attacks especially for logographic languages like Chinese. |
| Approach: | They propose a pretrained Chinese Bert that is robust to various forms of adversarial attacks like word perturbation, synonyms, typos, etc. |
| Outcome: | The proposed model outperforms baselines on 5 Chinese NLU tasks without sacrificing performance on clean testsets. |
Neural Data-to-Text Generation with LM-based Text Augmentation (2021.eacl-main)
Copied to clipboard
| Challenge: | Neural data-to-text generation is a difficult task for many new applications because of a lack of training data. |
| Approach: | They propose a few-shot approach that augments the data available for training by generating new text samples based on replacing specific values by alternative ones from the same category and pairing the new text with data samples. |
| Outcome: | The proposed approach outperforms fully supervised sequence-to-sequence models with less than 10% of the training set on both datasets. |
Diversifying Dialogue Generation with Non-Conversational Text (2020.acl-main)
Copied to clipboard
| Challenge: | Neural network-based sequence-to-sequence models suffer from low diversity in open-domain dialogue generation. |
| Approach: | They propose a way to diversify dialogue generation by leveraging non-conversational text . they collect large-scale corpus from forum comments, idioms and book snippets . |
| Outcome: | The proposed model produces significantly more diverse responses without sacrificing relevance with context. |
Multimodal Language Models See Better When They Look Shallower (2025.emnlp-main)
Copied to clipboard
Haoran Chen, Junyan Lin, Xinghao Chen, Yue Fan, Jianfeng Dong, Xin Jin, Hui Su, Jinlan Fu, Xiaoyu Shen
| Challenge: | Existing studies show that multimodal large language models extract visual features from the final layers of a pretrained Vision Transformer. |
| Approach: | They propose a feature fusion method that strategically incorporates shallower layers . they propose MLLMs that extract visual features from the final layers of a pretrained Vision Transformer . |
| Outcome: | The proposed method outperforms deep layers on fine-grained visual tasks . it is the first comprehensive study of visual layer selection for MLLMs . |
CSL: A Large-scale Chinese Scientific Literature Dataset (2022.coling-1)
Copied to clipboard
| Challenge: | Existing datasets centered around the English language restrict development of Chinese scientific NLP. |
| Approach: | They present a large-scale Chinese scientific literature dataset based on Chinese papers . they use semi-structured data as a natural annotation for many supervised NLP tasks . |
| Outcome: | The proposed dataset can serve as a Chinese corpus and perform many supervised tasks. |
MEIT: Multimodal Electrocardiogram Instruction Tuning on Large Language Models for Report Generation (2025.findings-acl)
Copied to clipboard
Zhongwei Wan, Che Liu, Xin Wang, Chaofan Tao, Hui Shen, Jing Xiong, Rossella Arcucci, Huaxiu Yao, Mi Zhang
| Challenge: | Recent studies have focused on classifying cardiac conditions using ECG data but have overlooked ECG report generation, which is time-consuming and requires clinical expertise. |
| Approach: | They propose a Multimodal ECG Instruction Tuning framework that extends the capability of large language models (LLMs) for the task. |
| Outcome: | The proposed framework outperforms open-source LLMs and LLM backbones across two large-scale ECG datasets. |
Argus: Benchmarking and Enhancing Vision-Language Models for 3D Radiology Report Generation (2025.findings-acl)
Copied to clipboard
| Challenge: | Existing work on 3D radiograph report generation focuses on 2D images, but 3D medical images provide more comprehensive diagnostic information. |
| Approach: | They propose a comprehensive training recipe for building high-performing VLMs for 3DRRG using a publicly available 3D CT-report dataset. |
| Outcome: | The proposed model achieves superior performance across different model sizes and input 3D medical image resolutions. |
How Controllable Are Large Language Models? A Unified Evaluation across Behavioral Granularities (2026.acl-long)
Copied to clipboard
Ziwen Xu, Kewei Xu, Haoming Xu, Haiwen Hong, Longtao Huang, Hui Xue, Ningyu Zhang, Yongliang Shen, Guozhou Zheng, Huajun Chen, Shumin Deng
| Challenge: | Large language models (LLMs) are increasingly deployed in socially sensitive domains, yet their unpredictable behaviors pose significant risks. |
| Approach: | They propose a hierarchical benchmark for evaluating LLM controllability across three domains: language features, sentiment, and personality. |
| Outcome: | The proposed framework offers a principled and interpretable framework for safe and controllable LLM behavior serving as a foundation for future research. |
MEDA: Dynamic KV Cache Allocation for Efficient Multimodal Long-Context Inference (2025.naacl-long)
Copied to clipboard
| Challenge: | Long-context Multimodal Large Language Models (MLLMs) require substantial computational resources as their multimodal Key-Value (KV) cache grows with increasing input lengths, challenging memory and time efficiency. |
| Approach: | They propose a dynamic multimodal KV cache allocation strategy that dynamically allocating KV size based on attention entropy to better adapt to multimodal interactions. |
| Outcome: | The proposed model achieves up to 72% KV cache memory reduction and 2.82 faster decoding speeds while maintaining or enhancing performance on various multimodal tasks in a long context. |
PEAR: Planner-Executor Agent Robustness Benchmark (2026.findings-eacl)
Copied to clipboard
| Challenge: | Existing studies examine isolated attack surfaces or specific scenarios, leaving a lack of holistic understanding of MAS vulnerabilities. |
| Approach: | They propose a benchmark to evaluate the utility and vulnerability of planner–executor MAS. |
| Outcome: | The proposed benchmark evaluates planner–executor MAS on a widely adopted design. |
AnchorMem: Anchored Facts with Associative Contexts for Building Memory in Large Language Models (2026.findings-acl)
Copied to clipboard
| Challenge: | Existing memory systems rely on summarization to preserve contextual nuances and obscuring key retrieval features. |
| Approach: | They propose a method that decouples the retrieval unit from the generation context. |
| Outcome: | The proposed method outperforms baseline models on the LoCoMo benchmark. |
Large Language Models are Effective Text Rankers with Pairwise Ranking Prompting (2024.findings-naacl)
Copied to clipboard
Zhen Qin, Rolf Jagerman, Kai Hui, Honglei Zhuang, Junru Wu, Le Yan, Jiaming Shen, Tianqi Liu, Jialu Liu, Donald Metzler, Xuanhui Wang, Michael Bendersky
| Challenge: | Existing methods to rank documents using large language models do not understand these challenging ranking formulations. |
| Approach: | They propose to use Pairwise Ranking Prompting to improve ranking performance . they propose to outperform fine-tuned baseline rankers on benchmark datasets . |
| Outcome: | The proposed technique outperforms supervised baselines on benchmark datasets and outperformed other LLM-based solutions by over 10% on average. |
AdaTP: Attention-Debiased Token Pruning for Video Large Language Models (2025.findings-emnlp)
Copied to clipboard
| Challenge: | Existing visual token compression methods rely on attention scores but have inherent biases . global and local attention biased scores cause excessive computational overhead . |
| Approach: | They propose a token pruning pipeline that targets global and local attention biases . the pipeline is designed to reduce computational overhead of Video Large Language Models based on visual tokens compiled from multiple video frames . |
| Outcome: | The proposed method significantly reduces the computational overhead of Video Large Language Models while retaining the performance of vanilla models. |
SVD-LLM V2: Optimizing Singular Value Truncation for Large Language Model Compression (2025.naacl-long)
Copied to clipboard
| Challenge: | Existing methods for compressing Large Language Models suffer from significant truncation losses. |
| Approach: | They propose a novel method that optimizes singular value truncation in SVD compression . they use dynamic compression ratio allocation to balance the large tuncation loss . |
| Outcome: | The proposed method outperforms current state-of-the-art methods on ten datasets and five models on various scales. |