Papers by Wenbin Zhang
FLAIR: Steering LLM Mathematical Problem Solving based on A Fuzzy-Logic-AssIsted Reasoner (2026.acl-long)
Copied to clipboard
Hao Wu, Hongru Sun, Wanqing Li, Xinguo Yu, Hao Ming, Xiao Luo, Wenbin Zhang, Jiahong Zhao, Yi Guo, Jie Yang
| Challenge: | Existing approaches to mathematical reasoning rely on static heuristics or pre-determined reasoning strategies. |
| Approach: | They propose an adaptive framework that integrates fuzzy theory into LLM-based mathematical reasoning. |
| Outcome: | The proposed framework outperforms state-of-the-art models while offering effective and interpretable diagnostics of intermediate problem-solving states. |
Hierarchical Representation-based Dynamic Reasoning Network for Biomedical Question Answering (2022.coling-1)
Copied to clipboard
Jianguo Mao, Jiyuan Zhang, Zengfeng Zeng, Weihua Peng, Wenbin Jiang, Xiangdong Wang, Hong Liu, Yajuan Lyu
| Challenge: | Existing models of biomedical question answering are limited in their ability to predict answers . a new model improves the performance of existing models, but the code will be released after the paper is published. |
| Approach: | They propose a hierarchical representation-based dynamic reasoning network to solve biomedical problems. |
| Outcome: | The proposed model significantly improves on three mainstream biomedical datasets . the code will be released after the paper is published . |
Mitigating the Alignment Tax of RLHF (2024.emnlp-main)
Copied to clipboard
Yong Lin, Hangyu Lin, Wei Xiong, Shizhe Diao, Jianmeng Liu, Jipeng Zhang, Rui Pan, Haoxiang Wang, Wenbin Hu, Hanning Zhang, Hanze Dong, Renjie Pi, Han Zhao, Nan Jiang, Heng Ji, Yuan Yao, Tong Zhang
| Challenge: | Large Language Models (LLMs) acquire a wide range of abilities during pre-training, but aligning LLMs under Reinforcement Learning with Human Feedback (RLHF) can lead to forgetting pretrained abilities, which is also known as the alignment tax. |
| Approach: | They propose to use a model averaging technique to find the most powerful alignment-forging Pareto front among RLHF algorithms. |
| Outcome: | The proposed method achieves the strongest alignment-forging Pareto front among competing methods. |
PDF-to-Tree: Parsing PDF Text Blocks into a Tree (2024.findings-emnlp)
Copied to clipboard
| Challenge: | Existing studies try to extract one universal reading order for PDF files, however, some applications, like Retrieval Augmented Generation, require breaking long articles into sections and subsections for better indexing. |
| Approach: | They propose a new task and dataset, PDF-to-Tree, which organizes the text blocks of a PDF into a tree structure. |
| Outcome: | The proposed parser achieves 93.93% accuracy, surpassing baseline methods by 6.72%. |
Characterizing the Impacts of Instances on Robustness (2023.findings-acl)
Copied to clipboard
Rui Zheng, Zhiheng Xi, Qin Liu, Wenbin Lai, Tao Gui, Qi Zhang, Xuanjing Huang, Jin Ma, Ying Shan, Weifeng Ge
| Challenge: | Existing defense approaches focus on developing new model structures or training algorithms, but they do little to tap the potential of training instances. |
| Approach: | They propose a method that can distinguish between robust and non-robust instances according to the model’s sensitivity to perturbations on individual instances during training. |
| Outcome: | The proposed method can distinguish between robust and non-robust instances according to the model’s sensitivity to perturbations on individual instances during training. |
Exploring the Vulnerability of the Content Moderation Guardrail in Large Language Models via Intent Manipulation (2025.findings-emnlp)
Copied to clipboard
| Challenge: | Prior work has shown that intent detection enhances LLMs’ moderation guardrails, but the robustness of these guardrail mechanisms under malicious manipulations remains under-explored. |
| Approach: | They propose a two-stage intent-based prompt-refinement framework that first transforms harmful inquiries into structured outlines and further reframes them into declarative-style narratives. |
| Outcome: | The proposed framework outperforms several cutting-edge jailbreak methods and evades even advanced Intent Analysis (IA) and Chain-of-Thought (CoT)-based defenses. |
Related Knowledge Perturbation Matters: Rethinking Multiple Pieces of Knowledge Editing in Same-Subject (2025.naacl-short)
Copied to clipboard
Zenghao Duan, Wenbin Duan, Zhiyi Yin, Yinghan Shen, Shaoling Jing, Jie Zhang, Huawei Shen, Xueqi Cheng
| Challenge: | Existing knowledge editing methods struggle when tasked with editing multiple related knowledge pieces for the same subject. |
| Approach: | They propose a benchmark to assess the effectiveness of knowledge editing methods . they use same-subject edits to ensure comprehensive updates to entity-centric knowledge . |
| Outcome: | The proposed method over-relys on subject information, neglecting other critical factors, resulting in reduced editing effectiveness. |
Variator: Accelerating Pre-trained Models with Plug-and-Play Compression Modules (2023.findings-emnlp)
Copied to clipboard
Chaojun Xiao, Yuqi Luo, Wenbin Zhang, Pengle Zhang, Xu Han, Yankai Lin, Zhengyan Zhang, Ruobing Xie, Zhiyuan Liu, Maosong Sun, Jie Zhou
| Challenge: | Large language models (LLMs) have been successful on NLP tasks but require huge parameter sizes and computational resources. |
| Approach: | They propose a parameter-efficient acceleration method that enhances computational efficiency through plug-and-play compression plugins. |
| Outcome: | The proposed method saves 53% computational costs using only 0.9% additional parameters with a performance drop of less than 2%. |
Learning to Conceal Risk: Controllable Multi-turn Red Teaming for LLMs in the Financial Domain (2026.acl-long)
Copied to clipboard
| Challenge: | Large Language Models (LLMs) are increasingly deployed in high-stakes domains such as finance where unsafe behavior can lead to serious regulatory risks. |
| Approach: | They propose a black-box multi-turn risk-concealed redteaming framework that progressively conceals surface-level risk while exploiting regulatory-violating behaviors. |
| Outcome: | Experiments on nine widely used LLMs show that the proposed framework achieves 93.19% average attack success rate (ASR) and improves the average ASR to 95.00%. |
Fine-Grained Legal Argument-Pair Extraction via Coarse-Grained Pre-training (2024.lrec-main)
Copied to clipboard
| Challenge: | Current methods conceptualize LAE as a supervised sentence-pair classification problem and necessitate extensive manual annotations. |
| Approach: | They propose a model that focuses on fine-grained alignment of argument pairs building upon coarse-grain complaint-defense pairs. |
| Outcome: | The proposed model outperforms baseline models by 3.7 and 2.4 points on average. |
Machine Reading Comprehension Using Structural Knowledge Graph-aware Network (D19-1)
Copied to clipboard
| Challenge: | Recent large-scale datasets specify that external knowledge is required to answer questions. |
| Approach: | They propose a model that leverages external knowledge to construct sub-graphs for entities in machine comprehension context. |
| Outcome: | The proposed model achieves state-of-the-art performance on the ReCoRD dataset. |
DynamicKV: Task-Aware Adaptive KV Cache Compression for Long Context LLMs (2025.findings-emnlp)
Copied to clipboard
| Challenge: | Existing KV cache compression methods enforce a fixed pattern, neglecting task-specific characteristics, which hampers the effective retention of essential information while discarding less important tokens. |
| Approach: | They propose a Task-Aware KV cache mechanism that dynamically adjusts the KV caching size across different layers based on the characteristics of the tasks. |
| Outcome: | The proposed method surpasses state-of-the-art methods by 11% on the LongBench dataset even under extreme compression (0.9%) |
Process-Supervised Reinforcement Learning for Code Generation (2025.emnlp-main)
Copied to clipboard
| Challenge: | Existing reinforcement learning strategies based on outcome supervision have shown effectiveness in code generation tasks, but their effectiveness in the field of code generation remains limited. |
| Approach: | They propose a method that uses a teacher model to mutate and refactor statements and a compiler to automatically label them. |
| Outcome: | The proposed method improves performance in complex code generation tasks. |