Papers by Fei Du
Steering Away from Refusal: A Black-box Jailbreak Method Based on First-Token Distribution (2026.findings-acl)
Copied to clipboard
| Challenge: | Existing methods to analyze black-box jailbreaks lack direct optimization signals to refine adversarial prompts. |
| Approach: | They propose a distribution-jailbreak attack method that selects effective jailbreak templates and iteratively optimizes adversarial suffixes by maximizing the KL divergence from the standard refusal distribution. |
| Outcome: | The proposed method achieves state-of-the-art Attack Success Rate (ASR) on all tested open-source models and delivers over 94% ASR on GPT-4.1. |
Model Composition for Multimodal Large Language Models (2024.acl-long)
Copied to clipboard
Chi Chen, Yiyang Du, Zheng Fang, Ziyue Wang, Fuwen Luo, Peng Li, Ming Yan, Ji Zhang, Fei Huang, Maosong Sun, Yang Liu
| Challenge: | Existing methods for creating versatile MLLMs rely on joint training with paired instruction data, which is resource-intensive and challenging to extend to new modalities. |
| Approach: | They propose a new paradigm for multimodal large language models by reusing modality encoders and merging LLM parameters. |
| Outcome: | The proposed model retains the modal understanding capabilities of each original model. |
LLMs Assist NLP Researchers: Critique Paper (Meta-)Reviewing (2024.emnlp-main)
Copied to clipboard
Jiangshu Du, Yibo Wang, Wenting Zhao, Zhongfen Deng, Shuaiqi Liu, Renze Lou, Henry Zou, Pranav Narayanan Venkit, Nan Zhang, Mukund Srinath, Haoran Zhang, Vipul Gupta, Yinghui Li, Tao Li, Fei Wang, Qin Liu, Tianlin Liu, Pengzhi Gao, Congying Xia, Chen Xing, Cheng Jiayang, Zhaowei Wang, Ying Su, Raj Shah, Ruohao Guo, Jing Gu, Haoran Li, Kangda Wei, Zihao Wang, Lu Cheng, Surangika Ranathunga, Meng Fang, Jie Fu, Fei Liu, Ruihong Huang, Eduardo Blanco, Yixin Cao, Rui Zhang, Philip Yu, Wenpeng Yin
| Challenge: | a comparative analysis of paper (meta-)reviews by large language models (LLMs) aims to identify and distinguish LLMs from human activities . |
| Approach: | They present a comparative analysis to identify and distinguish LLM activities from human activities. |
| Outcome: | The proposed analysis aims to improve recognition of instances when someone implicitly uses LLMs for reviewing activities. |
DualRAG: A Dual-Process Approach to Integrate Reasoning and Retrieval for Multi-Hop Question Answering (2025.acl-long)
Copied to clipboard
Rong Cheng, Jinyi Liu, Yan Zheng, Fei Ni, Jiazhen Du, Hangyu Mao, Fuzheng Zhang, Bo Wang, Jianye Hao
| Challenge: | Existing approaches to multi-hop question answering struggle to identify and organize dynamic knowledge . et al., 2023; Liu e.t. al. 2023) suggest a dual-process framework for multi-step reasoning . |
| Approach: | They propose a synergistic dual-process framework that integrates reasoning and retrieval. |
| Outcome: | The proposed framework improves answer accuracy and coherence even in smaller-scale models. |
PRDetect: Perturbation-Robust LLM-generated Text Detection Based on Syntax Tree (2025.findings-naacl)
Copied to clipboard
| Challenge: | Recent methods for detecting LLM-generated text have shown impressive performance, but in real-world scenarios, users often introduce perturbations to the text. |
| Approach: | They propose a method that detects syntactic trees that are minimally affected by perturbations and exhibit distinct differences between human-written and LLM-generated text. |
| Outcome: | The proposed method shows that it is significantly better against perturbations on the HC3 and GPT-3.5-mixed datasets and also has the shortest time expenditure. |
Second Language (Arabic) Acquisition of LLMs via Progressive Vocabulary Expansion (2025.acl-long)
Copied to clipboard
Jianqing Zhu, Huang Huang, Zhihang Lin, Juhao Liang, Zhengyang Tang, Khalid Almubarak, Mosen Alharthi, Bang An, Juncai He, Xiangbo Wu, Fei Yu, Junying Chen, Ma Zhuoheng, Yuhao Du, He Zhang, Saied Alshahrani, Emad A. Alghamdi, Lian Zhang, Ruoyu Sun, Haizhou Li, Benyou Wang, Jinchao Xu
| Challenge: | In the evolving landscape of large language models, the predominant focus has been on English and Chinese. |
| Approach: | They propose to utilize Arabic-specific vocabulary in the tokenizer to accelerate decoding. |
| Outcome: | The proposed model achieves decent performance comparable to the best Arabic LLMs across various Arabic benchmarks. |
Over-Generation and Compaction: A Prompting Strategy for Procedural Text Adaptation with Large Language Models (2025.findings-emnlp)
Copied to clipboard
| Challenge: | Existing prompting strategies for large language models often yield superficial or erroneous adaptations due to alignmentinduced biases and the inherent complexity of procedural editing. |
| Approach: | They propose an overgenerationandcompaction prompting strategy that leverages the model’s latent knowledge and compacts them into concise, coherent adaptations. |
| Outcome: | The proposed approach improves adaptation consistency and feasibility compared to baseline prompting methods without additional fine-tuning or curated training resources. |
Jailbreak LLMs through Internal Stance Manipulation (2025.emnlp-main)
Copied to clipboard
| Challenge: | Existing approaches to exploit LLMs' inherent safety mechanism, including GCG and AutoDAN, are ineffective for certain malicious requests. |
| Approach: | They propose a method that generates jailbreak prompts to suppress a refusal stance and induce affirmative responses by modifying adversarial prompts. |
| Outcome: | The proposed method outperforms the best baseline approach in Llama-2-7b-chat and achieves a 92.2% success rate across all models. |
Too Consistent to Detect: A Study of Self-Consistent Errors in LLMs (2025.emnlp-main)
Copied to clipboard
Hexiang Tan, Fei Sun, Sha Liu, Du Su, Qi Cao, Xin Chen, Jingang Wang, Xunliang Cai, Yuanzhuo Wang, Huawei Shen, Xueqi Cheng
| Challenge: | Existing detection methods fail to account for **self-consistent error** . study identifies self-consistency errors and evaluates them . |
| Approach: | They propose a method that fuses hidden state evidence from an external verifier LLM to detect self-consistent errors. |
| Outcome: | The proposed method significantly enhances performance on self-consistent errors across three LLM families. |
Conversational Question Answering with Language Models Generated Reformulations over Knowledge Graph (2024.findings-acl)
Copied to clipboard
| Challenge: | State-of-the-art ConvQA methods struggle with inexplicit question-answer pairs, which can degrade ConvQ performance. |
| Approach: | They propose a reinforcement learning based model, CoRnNet, which utilizes question reformulations generated by large language models to improve ConvQA performance. |
| Outcome: | The proposed model outperforms state-of-the-art ConvQA models by using question reformulations generated by large language models (LLMs). |
SCAIR: Schema-Conditioned Agentic Iterative Reasoning for Enterprise Knowledge Graphs (2026.acl-industry)
Copied to clipboard
Prateek Chaturvedi, Yuqicheng Zhu, Hongkuan Zhou, Dongzhuoran Zhou, Yunjie He, Steffen Staab, Fei Du, Jie Tang, Evgeny Kharlamov
| Challenge: | Existing agentic approaches for Knowledge Graph-based Retrieval-Augmented Generation fail to generalize to real-world enterprise Knowledge graphs (KGs) dense, schema-driven, and operationally constrained, requiring a training-free framework. |
| Approach: | They propose a training-free framework that integrates structured planning with controlled iterative reasoning by injecting schema-conditioned structural priors and enforcing schemas during multi-hop reasoning. |
| Outcome: | The proposed framework significantly improves on a real-world enterprise-oriented benchmark constructed from a Configuration Management DataBase (CMDB). |
Low-Entropy Watermark Detection via Bayes’ Rule Derived Detector (2025.findings-acl)
Copied to clipboard
| Challenge: | Existing methods for text watermarking ignore strong evidences embedded in low-entropy tokens, causing statistical measures to falsely indicate the absence of a watermark. |
| Approach: | They propose a Bayes' Rule derived watermark Detector which exploits watermark information from every token by leveraging the posterior probability of watermark’s presence. |
| Outcome: | The proposed method achieves 50% and 70% relative improvements over baselines in code generation and math problem-solving tasks. |
Step Potential Advantage Estimation: Harnessing Intermediate Confidence and Correctness for Efficient Mathematical Reasoning (2026.findings-acl)
Copied to clipboard
| Challenge: | Existing approaches to RLVR provide sparse supervision since reward arrives only after the full generation is complete. |
| Approach: | They propose a step-level reward system that extracts confidence and correctness and combines them into a Step Potential signal that explicitly estimates reasoning state at each step. |
| Outcome: | The proposed method outperforms existing methods on multiple benchmarks and improves accuracy while reducing response length. |
The Fall of ROME: Understanding the Collapse of LLMs in Model Editing (2024.findings-emnlp)
Copied to clipboard
| Challenge: | Recent studies have found that model editing methods can cause large language models to collapse with just a single edit. |
| Approach: | They propose a method that uses prefixed keys and adds prefixes during testing to prevent model collapse. |
| Outcome: | The proposed method prevents model collapse while maintaining effectiveness, the authors show . Rank-One Model Editing (ROME) has been found to cause model collapse with just a single edit . |
BaseCal: Unsupervised Confidence Calibration via Base Model Signals (2026.acl-long)
Copied to clipboard
Hexiang Tan, Wanli Yang, Junwei Zhang, Xin Chen, Rui Tang, Du Su, Jingang Wang, Yuanzhuo Wang, Fei Sun, Xueqi Cheng
| Challenge: | Post-trained LLMs typically compromise reliability with severe overconfidence, resulting in inaccurate responses. |
| Approach: | They propose a solution that feeds PoLLMs into the base LLM to get confidence. |
| Outcome: | The proposed solution reduces expected calibration error (ECE) by 42.90% compared to the best unsupervised baselines. |
ReSURE: Regularizing Supervision Unreliability for Multi-turn Dialogue Fine-tuning (2025.emnlp-main)
Copied to clipboard
| Challenge: | Existing methods address data quality via static prefiltering, which decouples quality control from training and fails to mitigate turn-level error propagation. |
| Approach: | They propose an adaptive learning method that dynamically down-weights unreliable supervision without explicit filtering. |
| Outcome: | Experiments on single-source and mixed-quality datasets show improved stability and response quality. |