Papers by Jinghui Chen
Stealthy and Persistent Unalignment on Large Language Models via Backdoor Injections (2024.naacl-long)
Copied to clipboard
| Challenge: | Recent advances in Large Language Models have demonstrated significant improvements in their capacity for generalization. |
| Approach: | They propose to conduct stealthy and persistent unalignment on large language models via backdoor injections. |
| Outcome: | The proposed approach passes the safety evaluation while maintaining strong persistence against re-alignment defense. |
Jailbreak Open-Sourced Large Language Models via Enforced Decoding (2024.acl-long)
Copied to clipboard
Hangfan Zhang, Zhimeng Guo, Huaisheng Zhu, Bochuan Cao, Lu Lin, Jinyuan Jia, Jinghui Chen, Dinghao Wu
| Challenge: | Existing studies show that Large Language Models can be misused to generate undesired content. |
| Approach: | They propose to use large language models to manipulate the generation process to generate undesired content without heavy computations or prompt designs. |
| Outcome: | The proposed method shows that open-sourced large language models could be misused to generate undesired content without heavy computations or prompt designs. |
ServImage: An Image Generation and Editing Benchmark from Real-world Commercial Imaging Services (2026.acl-long)
Copied to clipboard
Fengxian Ji, Jingpu Yang, Zirui Song, Lang Gao, Junhong Liang, Zhenhao Chen, Jinghui Zhang, Xiuying Chen
| Challenge: | Recent image generation and editing models demonstrate robust adherence to instructions and high visual quality on academic benchmarks. |
| Approach: | They propose a benchmark that correlates image outputs with economic value in commercial design projects. |
| Outcome: | ServImage benchmarks show image generation models perform well on academic benchmarks but are uncertain on commercial projects. |
Unbiased Math Word Problems Benchmark for Mitigating Solving Bias (2022.findings-naacl)
Copied to clipboard
| Challenge: | Existing solvers with data bias and learning bias only learn shallow heuristics rather than deep semantics for understanding problems. |
| Approach: | They propose a MWP dataset named UnbiasedMWP which is constructed by varying the grounded expressions in collected data and annotating them manually. |
| Outcome: | The proposed dataset has significantly fewer biases than its original data and other datasets, posing a promising benchmark for fairly evaluating the solvers’ reasoning skills rather than matching nearest neighbors. |
PlagBench: Exploring the Duality of Large Language Models in Plagiarism Generation and Detection (2025.naacl-long)
Copied to clipboard
| Challenge: | Recent studies have raised concerns about the potential threats large language models pose to academic integrity and copyright protection. |
| Approach: | They propose a dataset of 46.5K synthetic text pairs that represent three major types of plagiarism: verbatim copying, paraphrasing, and summarization. |
| Outcome: | The proposed dataset shows that GPT-3.5 Turbo can produce high-quality paraphrases and summaries without significantly increasing text complexity compared to GPT-4 Turbo. |
GeoQA: A Geometric Question Answering Benchmark Towards Multimodal Numerical Reasoning (2021.findings-acl)
Copied to clipboard
| Challenge: | Existing methods to solve geometric problems are dependent on handcraft rules and limited on small-scale datasets. |
| Approach: | They propose a Geometric Question Answering dataset with 5,010 geometric problems with corresponding annotated programs to illustrate the solving process. |
| Outcome: | The proposed method is significantly lower than human performance on the proposed dataset than on a publicly available dataset. |
JoPA: Explaining Large Language Model’s Generation via Joint Prompt Attribution (2025.acl-long)
Copied to clipboard
| Challenge: | Existing attempts to explain the entire language generation often treat input prompt texts independently, ignoring their combinatorial effects on the follow-up generation. |
| Approach: | They propose a framework for explaining how a few prompt texts collaboratively influences the LLM's complete generation. |
| Outcome: | The proposed explanations demonstrate faithfulness and efficiency of the proposed framework. |
FEDKIM: Adaptive Federated Knowledge Injection into Medical Foundation Models (2024.emnlp-main)
Copied to clipboard
| Challenge: | Existing foundation models are limited in access to diverse modalities and privacy regulations restrict the development of comprehensive foundation models. |
| Approach: | They propose a knowledge injection approach to extract and inject healthcare knowledge into medical foundation models to enhance their ability to handle multiple tasks and modalities. |
| Outcome: | The proposed method preserves privacy and enhances the model’s ability to handle complex medical tasks involving multiple modalities. |
WordGame: Efficient & Effective LLM Jailbreak via Simultaneous Obfuscation in Query and Response (2025.findings-naacl)
Copied to clipboard
| Challenge: | Recent advances in large language models have raised concerns about their susceptibility to jailbreaking attacks, which leads to harmful content inadvertently. |
| Approach: | They propose to exploit the safety alignment patterns of LLMs by simultaneous obfuscation in queries and responses to break down adversarial intent of query. |
| Outcome: | The proposed attack breaks down adversarial intent of query and encourages benign content regarding the games to precede anticipated harmful content in the response. |
Can Factual Opinions Be Edited (Manipulated) in Large Language Models? (2026.acl-long)
Copied to clipboard
| Challenge: | Existing methods for factual opinion editing focus on atomic facts, ignoring the risks associated with factual opinions. |
| Approach: | They propose a method that achieves opinion–evidence alignment without relying on explicit instructions to edit factual opinions. |
| Outcome: | The proposed method achieves opinion–evidence alignment without relying on explicit instructions. |
HyperText: Endowing FastText with Hyperbolic Geometry (2020.findings-emnlp)
Copied to clipboard
| Challenge: | Empirically, we show that HyperText outperforms FastText on a range of text classification tasks with much reduced parameters. |
| Approach: | They propose a model that uses hyperbolic geometry to model tree-like hierarchies in natural language sentences by embedding words or ngrams in hyperbolical space. |
| Outcome: | Empirically, the proposed model outperforms FastText on a range of text classification tasks with much reduced parameters. |
HyCoRec: Hypergraph-Enhanced Multi-Preference Learning for Alleviating Matthew Effect in Conversational Recommendation (2024.acl-long)
Copied to clipboard
| Challenge: | Existing methods to study the Matthew effect in Recommender Systems (RSs) however, it is amplified when the user interacts with the system over time. |
| Approach: | They propose a paradigm to alleviate the Matthew effect in conversational recommendation by learning multi-aspect preferences. |
| Outcome: | The proposed paradigm achieves state-of-the-art performance and superior of alleviating Matthew effect in conversational recommendation tasks. |
UniGeo: Unifying Geometry Logical Reasoning via Reformulating Mathematical Expression (2022.emnlp-main)
Copied to clipboard
| Challenge: | Existing work on geometry problem solving treats calculation and proving as two specific tasks hindering a deep model to unify reasoning ability on multiple math tasks. |
| Approach: | They propose a large-scale Unified Geometry problem benchmark to unify geometry on multiple math tasks. |
| Outcome: | The proposed framework outperforms the existing model with 5.6% and 3.2% accuracies on calculation and proving problems. |
Shadow-Activated Backdoor Attacks on Multimodal Large Language Models (2025.findings-acl)
Copied to clipboard
Ziyi Yin, Muchao Ye, Yuanpu Cao, Jiaqi Wang, Aofei Chang, Han Liu, Jinghui Chen, Ting Wang, Fenglong Ma
| Challenge: | Existing backdoor attacks on Multimodal Large Language Models are less applicable to open-ended conversations with users. |
| Approach: | They propose a shadow-activated backdoor attack scenario where attackers inject malicious content into the responses of MLLMs when the responses explicitly relate to the shadowed object. |
| Outcome: | The proposed framework achieves the desired behaviors by constructing a poisoned dataset and implementing an attention-regularized tuning strategy. |
PromptFix: Few-shot Backdoor Removal via Adversarial Prompt Tuning (2024.naacl-long)
Copied to clipboard
| Challenge: | Existing studies have shown that pre-trained language models can be backdoored such that model behavior is manipulated when trigger tokens are presented. |
| Approach: | They propose a backdoor mitigation strategy for NLP models via adversarial prompt-tuning in few-shot settings that uses two extra sets of soft tokens which approximate the trigger and counteract it respectively. |
| Outcome: | The proposed method keeps model parameters intact and only utilizes two extra sets of soft tokens which approximate the trigger and counteract it respectively. |
Defending Against Alignment-Breaking Attacks via Robustly Aligned LLM (2024.acl-long)
Copied to clipboard
| Challenge: | Large Language Models (LLMs) have made significant advancements but can be misused to generate harmful content. |
| Approach: | They propose a Robustly Aligned LLM to defend against alignment-breaking attacks by retraining existing LLMs and using adversarial or handcrafted jailbreaking prompts. |
| Outcome: | The proposed model reduces attack success rates from nearly 100% to around 10% or less. |
ICDAGENT: Empowering Agentic Large Language Models for Explainable Medical Coding (2026.acl-long)
Copied to clipboard
| Challenge: | Existing models lack convincing, human-understandable explanations, making them difficult for physicians to trust and use in practice. |
| Approach: | They propose a framework that aims to automatically assign ICD codes to clinical notes while providing explicit justifications for each assignment. |
| Outcome: | The proposed framework achieves effective ICD coding with accurate explanations using two collaborative LLM agents: a coding agent and a critical agent. |
MMLU-ProX: A Multilingual Benchmark for Advanced Large Language Model Evaluation (2025.emnlp-main)
Copied to clipboard
Weihao Xuan, Rui Yang, Heli Qi, Qingcheng Zeng, Yunze Xiao, Aosong Feng, Dairui Liu, Yun Xing, Junjue Wang, Fan Gao, Jinghui Lu, Yuang Jiang, Huitao Li, Xin Li, Kunyu Yu, Ruihai Dong, Shangding Gu, Yuekang Li, Xiaofei Xie, Felix Juefei-Xu, Foutse Khomh, Osamu Yoshie, Qingyu Chen, Douglas Teodoro, Nan Liu, Randy Goebel, Lei Ma, Edison Marrese-Taylor, Shijian Lu, Yusuke Iwasawa, Yutaka Matsuo, Irene Li
| Challenge: | Existing large language model evaluation benchmarks focus on English, while current multilingual tasks lack parallel questions that specifically assess cross-lingual reasoning abilities. |
| Approach: | They propose a comprehensive benchmark covering 29 languages, built on an English benchmark. |
| Outcome: | The MMLU-ProX is a comprehensive benchmark covering 29 languages, built on an English benchmark. |
When Personalization Tricks Detectors: The Feature-Inversion Trap in Machine-Generated Text Detection (2026.acl-long)
Copied to clipboard
Lang Gao, Xuhui Li, Chenxi Wang, Mingzhe Li, Wei Liu, Zirui Song, Jinghui Zhang, Rui Yan, Preslav Nakov, Xiuying Chen
| Challenge: | Personalized MGT detection remains largely underexplored due to personalization challenges . large language models (LLMs) can imitate personal writing styles, but they can generate fake news and misinformation. |
| Approach: | They propose a benchmark to evaluate detector robustness under personalization . they attribute this limitation to a feature-inversion trap that flips the effect in personalized contexts . |
| Outcome: | The proposed framework predicts detector robustness under personalization with an 85% correlation to actual results. |
Phi: Preference Hijacking in Multi-modal Large Language Models at Inference Time (2025.emnlp-main)
Copied to clipboard
| Challenge: | Recent advances in Multimodal Large Language Models have raised serious safety concerns. |
| Approach: | They propose a method for manipulating the output preference of MLLMs using a preference hijacked image. |
| Outcome: | The proposed method works at inference time and requires no model modifications. |
LogicSolver: Towards Interpretable Math Word Problem Solving with Logical Prompt-enhanced Learning (2022.findings-emnlp)
Copied to clipboard
| Challenge: | Recent advances in MWP solving are uninterpretable due to shallow heuristics . a new approach to solve automatic word problem solvers requires a solver to predict expression tree and corresponding linguistic logic formulas simultaneously. |
| Approach: | They propose to annotate interpretable logical formulas based on algebraic knowledge as the grounded linguistic logic of each solution equation. |
| Outcome: | The proposed approach improves interpretability of a MWP solver by using logical prompts and interpretation generation. |