Papers by Yujun Wang

ReCon: Active Defense against Large Vision-Language Model Jailbreaks via Reverse Safety Concept Injection (2026.findings-acl)

Challenge:	Large Reasoning Models (LRMs) often display unstable behaviors, e.g., hallucinating unsupported premises, overthinking simple tasks, and displaying higher sensitivity to prompt variations.
Approach:	They propose a graph-based analytical framework that clusters long, verbose CoT outputs into semantically coherent reasoning steps, then constructs directed reasoning graphs to capture contextual and logical dependencies among these steps.
Outcome:	The proposed framework enables quantitative evaluation of internal reasoning structure and quality beyond conventional metrics and provides practical insights for prompt engineering and cognitive analysis of LLMs.

Copied to clipboard

Zheng He, Yiwei Wang, Hongxing Wang, Yujun Cai

SLARD: A Chinese Superior Legal Article Retrieval Dataset (2025.coling-main)

Challenge:	Existing defense strategies neglect visual threats and lack of fine-grained specificity regarding specific attack semantics.
Approach:	They propose a black-box defense framework that maps unsafe concepts to fine-grained, constructive Safe Concepts.
Outcome:	a new black-box defense framework enhances robustness against jailbreak attacks . it maps detected unsafe concepts to fine-grained, constructive Safe Concepts . the proposed framework is available for free at http://www.epa.org/recon/ .

Copied to clipboard

Zhe Chen, Pengjie Ren, Fuhui Sun, Xiaoyan Wang, Yujun Li, Siwen Zhao, Tengyi Yang

Information Retrieval

Challenge:	Existing retrieval methods struggle to achieve ideal results, a study finds . existing large language models lack prior knowledge of the content of superior legal articles .
Approach:	They propose to use a Chinese superior legal article retrieval dataset to find relevant articles with higher legal effectiveness.
Outcome:	The proposed dataset shows that existing retrieval methods struggle to achieve ideal results.

Understanding GUI Agent Localization Biases through Logit Sharpness (2025.findings-emnlp)

Copied to clipboard

Xingjian Tao, Yiwei Wang, Yujun Cai, Zhicheng Yang, Jing Tang

Challenge:	Multimodal large language models often exhibit hallucinations that compromise reliability . despite promising performance, these models often display systematic localization errors .
Approach:	They propose a framework that categorizes model predictions into four distinct types . they propose metric that evaluates alignment between semantic continuity and logits distribution .
Outcome:	The proposed framework categorizes model predictions into four different types . it reveals nuanced failure modes beyond traditional accuracy metrics .

LLaVA Steering: Visual Instruction Tuning with 500x Fewer Parameters through Modality Linear Representation-Steering (2025.acl-long)

Copied to clipboard

Jinhe Bi, Yujun Wang, Haokun Chen, Xun Xiao, Artur Hecker, Volker Tresp, Yunpu Ma

Semantic Text Processing

Challenge:	Multimodal Large Language Models (MLLMs) enhance visual tasks by integrating visual representations into large language models.
Approach:	They propose a method to re-balance modalities by steering visual representations . they propose LLaVA Steering, a platform that enables rapid customization of MLLMs a component-based architecture .
Outcome:	The proposed model re-balances the modalities of visual representations in large language models . the model requires 500 times fewer trainable parameters than LoRA while maintaining comparable performance .

GENUINE: Graph Enhanced Multi-level Uncertainty Estimation for Large Language Models (2025.findings-emnlp)

Copied to clipboard

Tuo Wang, Adithya Kulkarni, Tyler Cody, Peter A. Beling, Yujun Yan, Dawei Zhou

Primacy Effect of ChatGPT (2023.emnlp-main)

Challenge:	Existing methods for estimation of uncertainty overlook semantic dependencies, authors say . genUINE: Graph ENhanced mUlti-level uncertainty Estimation for Large Language Models leverages dependency parse trees and hierarchical graph pooling .
Approach:	They propose a graph-enhanced mUlti-level uncertaINty estimation framework that leverages dependency parse trees and hierarchical graph pooling to refine uncertainty quantification.
Outcome:	The proposed framework achieves higher AUROC and lower calibration errors than existing methods.

Copied to clipboard

Yiwei Wang, Yujun Cai, Muhao Chen, Yuxuan Liang, Bryan Hooi

VistaWise: Building Cost-Effective Agent with Cross-Modal Knowledge Graph for Minecraft (2025.emnlp-main)

Challenge:	Existing machine learning models may lead to poor performance in discriminative natural language understanding tasks.
Approach:	They propose to use ChatGPT to query large amounts of human-written text to find the answer to a question.
Outcome:	The proposed model has a high chance to select labels at earlier positions as the answer.

Copied to clipboard

Honghao Fu, Junlong Ren, Qi Chai, Deheng Ye, Yujun Cai, Hao Wang

Should We Rely on Entity Mentions for Relation Extraction? Debiasing Relation Extraction with Counterfactual Analysis (2022.naacl-main)

Challenge:	Large language models (LLMs) have shown significant promise in embodied decision-making tasks within virtual open-world environments, but lack domain-specific knowledge.
Approach:	They propose a cost-effective agent framework that integrates cross-modal domain knowledge and finetunes a dedicated object detection model for visual analysis.
Outcome:	The proposed framework reduces the requirement for domain-specific training data from millions of samples to a few hundred.

Copied to clipboard

Yiwei Wang, Muhao Chen, Wenxuan Zhou, Yujun Cai, Yuxuan Liang, Dayiheng Liu, Baosong Yang, Juncheng Liu, Bryan Hooi

Information Extraction & Text Mining

Challenge:	Existing studies rely on entity information for sentence-level relation extraction (RE) but this can leak superficial and spurious clues of relations.
Approach:	They propose to use entity mentions to extract relations from textual context . they use a causal graph to model dependencies between variables in RE models .
Outcome:	The proposed method yields significant gains on both effectiveness and generalization for RE.

MRFD: Multi-Region Fusion Decoding with Self-Consistency for Mitigating Hallucinations in LVLMs (2025.findings-emnlp)

Copied to clipboard

Haonan Ge, Yiwei Wang, Ming-Hsuan Yang, Yujun Cai

Reasoning

Challenge:	Large Vision-Language Models often produce hallucinations due to the limited ability to verify information in different regions of the image.
Approach:	a new decoding method improves factual grounding by modeling inter-region consistency . the method identifies salient regions using cross-attention and generates initial responses for each .
Outcome:	a training-free decoding method reduces hallucinations and improves response consistency . the proposed method generates initial responses for each region and weights reliability weights among responses .

SemVink: Advancing VLMs’ Semantic Understanding of Optical Illusions via Visual Global Thinking (2025.emnlp-main)

Copied to clipboard

Sifan Li, Yujun Cai, Yiwei Wang

Visual Data in NLP

Challenge:	Vision-language models excel in semantic tasks but fail at detecting hidden content . current architectures prioritize abstract reasoning over low-level visual operations .
Approach:	They propose a benchmark to test vision-language models that can detect hidden content . they propose HC-Bench to scale images to low resolutions to unlock 99% accuracy .
Outcome:	HC-Bench shows that leading VLMs achieve near-zero accuracy even with explicit prompting . et al.: current models prioritize abstract reasoning over low-level visual operations . they urge a shift toward hybrid models bridging gap between computational vision and human cognition .

Representation Interventions Enable Lifelong Knowledge Memory Control in LLMs (2026.acl-long)

Copied to clipboard

Xuyuan Liu, Shengyu Chen, Xinshuai Dong, Yanchi Liu, Xujiang Zhao, Haoyu Wang, Yujun Yan, Haifeng Chen, Zhengzhang Chen

Semantic Text Processing

Challenge:	Large language models (LLMs) produce outdated or inaccurate content. Updating their knowledge efficiently and accurately without costly retraining is a major challenge.
Approach:	They propose a robust and scalable method that treats knowledge control as interventions within the model’s representation space.
Outcome:	The proposed method achieves fine-grained control over complex, unstructured knowledge while maintaining general utility with frozen base weights.

A Structure-Aware Argument Encoder for Literature Discourse Analysis (2022.coling-1)

Copied to clipboard

Yinzi Li, Wei Chen, Zhongyu Wei, Yujun Huang, Chujun Wang, Siyuan Wang, Qi Zhang, Xuanjing Huang, Libo Wu

Semantic Text Processing

Challenge:	Existing research for argument representation learning treats tokens in sentences equally and ignores the implied structure information of argumentative context.
Approach:	They propose to separate tokens into two groups to capture structural information of arguments and to incorporate paragraph-level position information into the model.
Outcome:	The proposed model captures structural information of arguments and is able to identify arguments automatically.

Vulnerability of LLMs to Vertically Aligned Text Manipulations (2025.acl-long)

Copied to clipboard

Zhecheng Li, Yiwei Wang, Bryan Hooi, Yujun Cai, Zhen Xiong, Nanyun Peng, Kai-Wei Chang

Text Classification

Challenge:	Recent research shows that vertical text input significantly degrades the accuracy of large language models (LLMs) in text classification tasks.
Approach:	They investigate the impact of vertical text input on the performance of LLMs . they find that chain of thought reasoning does not help LLM recognize vertical input .
Outcome:	The proposed model can significantly mislead models, posing a risk of bypassing detection in real-world scenarios involving harmful or sensitive information.

Dynamic Infilling Anchors for Format-Constrained Generation in Diffusion Large Language Models (2026.acl-long)

Copied to clipboard

Boyan Han, Yiwei Wang, Yi Song, Yujun Cai, Chi Zhang

Learning to Detect Noisy Labels Using Model-Based Features (2022.findings-emnlp)

Challenge:	Diffusion large language models (dLLMs) offer bidirectional attention and parallel generation . fixed anchors can enforce constraints, but they often impose rigid spans, leading to truncated reasoning .
Approach:	They propose a method that dynamically estimates end-anchor positions to adjust generation length before iterative infilling.
Outcome:	The proposed method improves format compliance and answer accuracy on GSM8K and MATH.

Copied to clipboard

Zhihao Wang, Zongyu Lin, Junjie Wen, Xianxin Chen, Peiqi Liu, Guidong Zheng, Yujun Chen, Zhilin Yang

DT-Solver: Automated Theorem Proving with Dynamic-Tree Sampling Guided by Proof-level Value Function (2023.acl-long)

Challenge:	Existing approaches to reduce label noise rely on heuristics and sample losses.
Approach:	They propose a method that transfers the noise distribution to a clean set and trains a model to distinguish noisy labels from clean ones using model-based features.
Outcome:	Empirically, the proposed approach improves over strong baselines on a wide range of tasks including text classification and speech recognition.

Copied to clipboard

Haiming Wang, Ye Yuan, Zhengying Liu, Jianhao Shen, Yichun Yin, Jing Xiong, Enze Xie, Han Shi, Yujun Li, Lin Li, Jian Yin, Zhenguo Li, Xiaodan Liang

PolicyLLM: Towards Excellent Comprehension of Public Policy for Large Language Models (2026.findings-acl)

Challenge:	Recent advances in neural theorem-proving resort to large language models and tree searches.
Approach:	They propose a Dynamic-Tree Driven Theorem Solver to accommodate general theoremes by guiding the search procedure with state confidence and proof-level values.
Outcome:	The proposed method outperforms state-of-the-art methods on two popular theorem-proving datasets with a 6.65% improvement on average in terms of success rate.

Copied to clipboard

Han Bao, Penghao Zhang, Yue Huang, Zhengqing Yuan, Yanchi Ru, SU Rui, Yujun Zhou, Xiangqi Wang, Kehan Guo, Nitesh V Chawla, Yanfang Ye, Xiangliang Zhang

Tricking Retrievers with Influential Tokens: An Efficient Black-Box Corpus Poisoning Attack (2025.naacl-long)

Challenge:	Large Language Models (LLMs) are increasingly integrated into real-world decision-making, but their ability to comprehend and reason about policy-related content remains underexplored.
Approach:	They propose a bilingual benchmark evaluating policy comprehension comprising 21K cases across a broad spectrum of policy areas.
Outcome:	The proposed model shows stronger performance on application-oriented policy tasks than on memorization or conceptual understanding, and yields the highest accuracy on structured reasoning tasks.

Copied to clipboard

Cheng Wang, Yiwei Wang, Yujun Cai, Bryan Hooi

Con-ReCall: Detecting Pre-training Data in LLMs via Contrastive Decoding (2025.coling-main)

Challenge:	Existing methods for crafting adversarial passages are slow and computationally expensive, requiring either access to retriever’s gradients or large computational resources.
Approach:	They propose a method that leverages two key characteristics of retrievers: insensitivity to token order and bias towards influential tokens to generate effective adversarial passages.
Outcome:	The proposed method achieves superior efficiency and scalability compared to existing methods while maintaining comparable or better attack success rates across multiple datasets.

Copied to clipboard

Cheng Wang, Yiwei Wang, Bryan Hooi, Yujun Cai, Nanyun Peng, Kai-Wei Chang

Dissecting Logical Reasoning in LLMs: A Fine-Grained Evaluation and Supervision Study (2025.findings-emnlp)

Challenge:	Existing methods analyze training data with member and non-member contexts, overlooking potential insights from both member and not-member.
Approach:	They propose a method that leverages asymmetric distributional shifts induced by member and non-member contexts through contrastive decoding to enhance membership inference.
Outcome:	The proposed approach outperforms the current state-of-the-art on the WikiMIA benchmark and is robust against various text manipulation techniques.

Copied to clipboard

Yujun Zhou, Jiayi Ye, Zipeng Ling, Yufei Han, Yue Huang, Haomin Zhuang, Zhenwen Liang, Kehan Guo, Taicheng Guo, Xiangqi Wang, Xiangliang Zhang

Making Every Step Effective: Jailbreaking Large Vision-Language Models Through Hierarchical KV Equalization (2025.findings-emnlp)

Challenge:	Existing benchmarks that rely on final-answer accuracy fail to capture the quality of the reasoning process.
Approach:	They propose a fine-grained evaluation framework that assesses logical reasoning across three dimensions: overall accuracy, stepwise soundness, and representation-level probing.
Outcome:	The proposed framework assesses logical reasoning across three dimensions: overall accuracy, stepwise soundness, and representation-level probing.

Copied to clipboard

Shuyang Hao, Yiwei Wang, Bryan Hooi, Jun Liu, Muhao Chen, Zi Huang, Yujun Cai

Robustness in NLP

Challenge:	HKVE selectively accepts gradient optimization results based on the distribution of attention scores across different layers, ensuring that every optimization step positively contributes to the attack.
Approach:	They propose a framework that selectively accepts gradient optimization results based on the distribution of attention scores across different layers and selectively takes them into account when calculating the attack success rate.
Outcome:	The proposed framework outperforms existing methods by achieving success rates of 75.08% on MiniGPT4, 85.84% on LLaVA and 81.00% on Qwen-VL.

DiMo-GUI: Advancing Test-time Scaling in GUI Grounding via Modality-Aware Visual Reasoning (2025.emnlp-main)

Copied to clipboard

Hang Wu, Hongkai Chen, Yujun Cai, Chang Liu, Qingwen Ye, Ming-Hsuan Yang, Yiwei Wang

DRS: Deep Question Reformulation With Structured Output (2025.findings-acl)

Challenge:	DiMo-GUI is a training-free framework for GUI grounding that splits input into textual elements and iconic elements, allowing the model to reason over each modality independently using general-purpose vision-language models.
Approach:	They propose a training-free framework for GUI grounding that leverages two core strategies: dynamic visual grounding and modality-aware optimization.
Outcome:	The proposed framework splits the input into textual elements and iconic elements, allowing the model to reason over each modality independently using general-purpose vision-language models.

Copied to clipboard

Zhecheng Li, Yiwei Wang, Bryan Hooi, Yujun Cai, Nanyun Peng, Kai-Wei Chang

GPS: Genetic Prompt Search for Efficient Few-Shot Learning (2022.emnlp-main)

Challenge:	Existing models like GPT-3 and Instruct-GPT lack the ability to reformulate unanswerable questions.
Approach:	They propose a zero-shot method that combines the strengths of LLMs with a DFS-based algorithm to iteratively explore potential entity combinations and constrain outputs using predefined entities.
Outcome:	The proposed method outperforms all baselines, including the GPT-3.5 model, on the unanswerable question reformulation task.

Copied to clipboard

Hanwei Xu, Yujun Chen, Yulun Du, Nan Shao, Wang Yanggang, Haiyu Li, Zhilin Yang

K-GIP: Diagnosing Logical Fractures in Large Vision-Language Models via Verification Scene Graphs and Sequential Pruning (2026.findings-acl)

Challenge:	Pretrained language models are often finetuned for downstream tasks, which has been shown to improve performance over non-pretrained models.
Approach:	They propose a genetic algorithm to automatically search for the best prompt for few-shot learning with pretrained language models by gradient-free algorithm.
Outcome:	Experiments on diverse datasets show that the proposed method outperforms manual prompts by 2.6 points.

Copied to clipboard

Yujun Hu, Xiaoyu Zhou, Changbo Wang, Gaoqi He

Graph-Assisted Large Language Models: A Perspective on Mitigating Intrinsic Limitations (2026.findings-acl)

Challenge:	Existing benchmarks that treat hallucinations as isolated errors neglect causal dependencies between visual perception and textual reasoning.
Approach:	They propose a Knowledge-Guided In-Context Probing framework that constructs a dual-perception ground truth to transform abstract priors into multi-granularity queries.
Outcome:	The proposed framework isolates deep reasoning failures from simple perceptual misses.

Copied to clipboard

Haitong Luo, Fali Wang, Weiyao Zhang, Xianren Zhang, Zhiwei Zhang, Tianxiang Zhao, Minhua Lin, Jiahao Zhang, Hui Liu, Xianfeng Tang, Qi He, Suhang Wang, Xuying Meng, Yujun Zhang

Structured Data in NLP

Challenge:	Large language models exhibit intrinsic limitations such as knowledge cutoff, single-threaded reasoning that hinders finer-grained branch and aggregation, and rigid collaboration mechanisms that struggle to coordinate specialized capabilities.
Approach:	They propose a taxonomy spanning Graph-Assisted Knowledge Augmentation, Graph Assisted Reasoning and Planning, and Graphed LLM Collaboration.
Outcome:	The proposed models show that graphs can augment and correct LLMs and support dynamic coordination among experts and agents in collaborative settings.

Mitigating Coordinate Prediction Bias from Positional Encoding Failures (2026.findings-acl)

Copied to clipboard

Xingjian Tao, Yiwei Wang, Yujun Cai, Yihong Luo, Kai Han, Jing Tang

ZeroPrompt: Scaling Prompt-Based Pretraining to 1,000 Tasks Improves Zero-Shot Generalization (2022.findings-emnlp)

Challenge:	Multimodal Large Language Models (MLLMs) excel at general vision-language tasks, but precise coordinate prediction remains a challenge.
Approach:	They propose a training-free, inference-time correction method to correct VPEs . they isolate position-unconditioned tendencies by shuffling VPE and use it to steer digit decoding .
Outcome:	The proposed method is training-free, inference-time correction method . it effectively rectifies coordinate drift, yielding consistent improvements without retraining .

Copied to clipboard

Hanwei Xu, Yujun Chen, Yulun Du, Nan Shao, Wang Yanggang, Haiyu Li, Zhilin Yang