Papers by Yiwei Wang

BlindGuard: Safeguarding LLM-based Multi-Agent Systems under Unknown Attacks (2026.acl-long)

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

Rui Miao, Yixin Liu, Yili Wang, Xu Shen, Yue Tan, Yiwei Dai, Shirui Pan, Xin Wang

Robustness in NLP

Challenge:	Existing supervised defense methods rely on labeled malicious agents to train a supervised model of malicious behavior.
Approach:	They propose an unsupervised defense method that learns without requiring any attack-specific labels or prior knowledge of malicious behaviors.
Outcome:	The proposed method detects diverse attack types across MAS with various communication patterns while maintaining superior generalizability compared to baselines.

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

Copied to clipboard

Zheng He, Yiwei Wang, Hongxing Wang, Yujun Cai

LLM-Powered Benchmark Factory: Reliable, Generic, and Efficient (2026.acl-long)

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

Peiwen Yuan, Shaoxiong Feng, Yiwei Li, Xinglin Wang, Yueqi Zhang, Jiayi Shi, Chuyi Tan, Boyuan Pan, Yao Hu, Kan Li

Context-DPO: Aligning Language Models for Context-Faithfulness (2025.findings-acl)

Challenge:	Using generic and efficient benchmark generators, human annotators are limited by inefficiency . current benchmark generator methods rely on seed signals, leading to long cycles and high costs .
Approach:	They propose a framework to evaluate LLMs as generic benchmark generators and integrate them as BenchMaker.
Outcome:	The proposed framework achieves comparable performance to human-annotated benchmarks on most metrics.

Copied to clipboard

Baolong Bi, Shaohan Huang, Yiwei Wang, Tianchi Yang, Zihan Zhang, Haizhen Huang, Lingrui Mei, Junfeng Fang, Zehao Li, Furu Wei, Weiwei Deng, Feng Sun, Qi Zhang, Shenghua Liu

Decoding by Contrasting Knowledge: Enhancing Large Language Model Confidence on Edited Facts (2025.acl-long)

Challenge:	Context-DPO is the first alignment method specifically designed to enhance contextfaithfulness for large language models.
Approach:	They propose a benchmark that simulates Retrieval-Augmented Generation scenarios with knowledge conflicts to evaluate context-faithfulness.
Outcome:	The proposed method improves LLMs' context-faithfulness by 35% to 280% over open-source models.

Copied to clipboard

Baolong Bi, Shenghua Liu, Lingrui Mei, Yiwei Wang, Junfeng Fang, Pengliang Ji, Xueqi Cheng

Make Every Penny Count: Difficulty-Adaptive Self-Consistency for Cost-Efficient Reasoning (2025.findings-naacl)

Challenge:	In-context knowledge editing (ICE) is currently the most effective method for knowledge editing, but it is constrained by the black-box modeling of LLMs and lacks interpretability.
Approach:	They propose a method to decode new knowledge by comparing logits with unedited knowledge to improve the accuracy of LLMs.
Outcome:	The proposed method improves the performance of LLaMA3-8B-instruct on MQuAKE by up to 219%.

Copied to clipboard

Xinglin Wang, Shaoxiong Feng, Yiwei Li, Peiwen Yuan, Yueqi Zhang, Chuyi Tan, Boyuan Pan, Yao Hu, Kan Li

Gated Differentiable Working Memory for Long-Context Language Modeling (2026.acl-long)

Challenge:	Existing decoding strategies for chain-of-thought reasoning do not exploit prior information about question difficulty.
Approach:	They propose a decoding strategy called self-consistency to improve reasoning performance by adjusting the number of samples based on the posterior distribution of a set of pre-samples.
Outcome:	The proposed method outperforms baseline methods on arithmetic, commonsense and symbolic reasoning tasks while achieving comparable performance.

Copied to clipboard

Lingrui Mei, Shenghua Liu, Yiwei Wang, Yuyao Ge, Baolong Bi, Jiayu Yao, Jun Wan, Ziling Yin, Jiafeng Guo, Xueqi Cheng

LLM-A*: Large Language Model Enhanced Incremental Heuristic Search on Path Planning (2024.findings-emnlp)

Challenge:	Long contexts break transformers, attention scores dilute, model cannot adapt to novel patterns at inference time.
Approach:	They propose a framework that gates the memory consolidation process by estimating Contextual Utility . they propose GDWM to maintain a form of working memory with constant contexts .
Outcome:	The proposed framework achieves comparable or superior performance on sparse-information tasks with 4 fewer gradient steps compared to uniform baselines.

Copied to clipboard

Silin Meng, Yiwei Wang, Cheng-Fu Yang, Nanyun Peng, Kai-Wei Chang

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

Challenge:	Existing path planning algorithms suffer from significant computational and memory inefficiencies as the state space grows . large language models excel in environmental analysis but fall short in detailed spatial and temporal reasoning .
Approach:	They propose a new path planning method that synergistically combines A* and LLMs to improve pathfinding efficiency.
Outcome:	The proposed method improves pathfinding efficiency while maintaining integrity of path validity in large-scale scenarios.

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 .

SPARD: Self-Paced Curriculum for RL Alignment via Integrating Reward Dynamics and Data Utility (2026.acl-long)

Copied to clipboard

Xuyang Zhi, Peilun Zhou, Chengqiang Lu, Hang Lv, Yiwei Liang, Rongyang Zhang, Yan Gao, null Yiwu, Yao Hu, Hongchao Gu, Defu Lian, Hao Wang, Enhong Chen

Challenge:	Large Language Models (LLMs) are shifting the focus from single verifiable tasks toward complex, open-ended real-world scenarios.
Approach:	They propose a framework that automatically adjusts reward weights and data importance to synchronize learning intent with data utility for optimal performance.
Outcome:	The proposed framework improves model capabilities across all domains and scales.

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

Explainability & Interpretability in NLP

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.

Mitigate Extrinsic Social Bias in Pre-trained Language Models via Continuous Prompts Adjustment (2024.emnlp-main)

Copied to clipboard

Yiwei Dai, Hengrui Gu, Ying Wang, Xin Wang

LPNL: Scalable Link Prediction with Large Language Models (2024.findings-acl)

Challenge:	Existing methods of extrinsic bias mitigation rely on manual word lists for sensitive groups . however, these word lists are limited by length and scope, resulting in poor performance.
Approach:	They propose a method which generates continuous token lists from the entire vocabulary space and uses them to bridge the gap between outputs and targets in fairness learning process.
Outcome:	The proposed method outperforms baseline methods on three NLU tasks.

Copied to clipboard

Baolong Bi, Shenghua Liu, Yiwei Wang, Lingrui Mei, Xueqi Cheng

Primacy Effect of ChatGPT (2023.emnlp-main)

Challenge:	Existing studies on graph learning with large language models have focused on the link prediction task on large graphs.
Approach:	They propose a framework for scalable link prediction on large-scale heterogeneous graphs based on large language models.
Outcome:	The proposed framework outperforms baselines in link prediction tasks on large graphs.

Copied to clipboard

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

Focused Large Language Models are Stable Many-Shot Learners (2024.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

Peiwen Yuan, Shaoxiong Feng, Yiwei Li, Xinglin Wang, Yueqi Zhang, Chuyi Tan, Boyuan Pan, Heda Wang, Yao Hu, Kan Li

Control Large Language Models via Divide and Conquer (2024.emnlp-main)

Challenge:	In-Context Learning (ICL) enables large language models to achieve rapid task adaptation by learning from demonstrations.
Approach:	They propose a training-free method that disperses model attention from the query . they propose 'focus' search strategy that uses model perplexity to ensure sufficient attention .
Outcome:	The proposed method achieves an average performance improvement of 5.2% over vanilla ICL and scales well with many-shot demonstrations.

Copied to clipboard

Bingxuan Li, Yiwei Wang, Tao Meng, Kai-Wei Chang, Nanyun Peng

InsBank: Evolving Instruction Subset for Ongoing Alignment (2025.findings-emnlp)

Challenge:	Lexically Constrained Generation (LCG) is a crucial task of text generation.
Approach:	They propose a Divide and Conquer Generation strategy to enhance LLMs' performance in Lexically Constrained Generation with prompt-based controlling.
Outcome:	The proposed strategy shows 90% improvement on the most challenging LCG task.

Copied to clipboard

Jiayi Shi, Yiwei Li, Shaoxiong Feng, Peiwen Yuan, Xinglin Wang, Yueqi Zhang, Chuyi Tan, Boyuan Pan, Huan Ren, Yao Hu, Kan Li

Green & Sustainable NLP

Challenge:	Recent studies emphasize that quality and diversity of instruction data are more crucial than quantity, highlighting the need to select diverse, high-quality subsets to reduce training costs.
Approach:	They propose to use a continuously updated repository to integrate the latest valuable instruction data with a progressive evolution framework to evolve InsBank over time.
Outcome:	The proposed framework outperforms baselines in InsBank evolution and extracts budget-specific subsets.

A Causal View of Entity Bias in (Large) Language Models (2023.findings-emnlp)

Copied to clipboard

Fei Wang, Wenjie Mo, Yiwei Wang, Wenxuan Zhou, Muhao Chen

Integrate the Essence and Eliminate the Dross: Fine-Grained Self-Consistency for Free-Form Language Generation (2024.acl-long)

Challenge:	Entity bias affects pretrained (large) language models, causing them to rely on (biased) parametric knowledge to make unfaithful predictions.
Approach:	They propose a structured causal model whose parameters are easier to estimate . they propose to perturb the original entity with neighboring entities .
Outcome:	The proposed model reduces biasing information pertaining to the original entity while still preserving sufficient semantic information from similar entities.

Copied to clipboard

Xinglin Wang, Yiwei Li, Shaoxiong Feng, Peiwen Yuan, Boyuan Pan, Heda Wang, Yao Hu, Kan Li

Modeling Complex Dialogue Mappings via Sentence Semantic Segmentation Guided Conditional Variational Auto-Encoder (2022.findings-emnlp)

Challenge:	Existing methods to improve output quality without aggregating input tokens are limited by the complexity of aggregation of responses.
Approach:	They propose to extract and integrate segment-level commonalities from candidate samples to enhance performance of LLMs in open-ended and reasoning tasks.
Outcome:	The proposed method improves performance on reasoning, code generation and mathematical reasoning tasks without requiring additional models and overlooking the knowledge present among the candidates.

Copied to clipboard

Bin Sun, Shaoxiong Feng, Yiwei Li, Weichao Wang, Fei Mi, Yitong Li, Kan Li

Dialogue Systems & Conversational Agents

Challenge:	Existing efforts to identify and avoid CDM to facilitate dialogue learning failed to solve the problem.
Approach:	They propose a Sentence Semantic Segmentation guided Conditional Variational Auto-Encoder which can model and take advantage of the CDM data.
Outcome:	The proposed method can model and take advantages of the CDM data.

Tackling Modality Heterogeneity with Multi-View Calibration Network for Multimodal Sentiment Detection (2023.acl-long)

Copied to clipboard

Yiwei Wei, Shaozu Yuan, Ruosong Yang, Lei Shen, Zhangmeizhi Li, Longbiao Wang, Meng Chen

Dangling-Aware Entity Alignment with Mixed High-Order Proximities (2022.findings-naacl)

Challenge:	Existing studies focus on fusing different features but ignore the challenge of modality heterogeneity.
Approach:	They propose a text-guided fusion module with novel Sparse-Attention to reduce the negative impacts of redundant visual elements and a sentiment-based congruity constraint task to calibrate the feature shift in the representation space.
Outcome:	The proposed model is competitive against existing methods and achieves state-of-the-art results on two public benchmark datasets.

Copied to clipboard

Juncheng Liu, Zequn Sun, Bryan Hooi, Yiwei Wang, Dayiheng Liu, Baosong Yang, Xiaokui Xiao, Muhao Chen

Structured Data in NLP

Challenge:	Existing methods for dangling-aware entity alignment are underexplored but important problem.
Approach:	They propose a framework that uses high-order proximities to detect dangling entities and align matchable entities.
Outcome:	The proposed framework detects dangling entities and aligns matchable entities better than existing methods.

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

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

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

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 .

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 .

a1: Steep Test-time Scaling Law via Environment Augmented Generation (2026.findings-acl)

Copied to clipboard

Lingrui Mei, Shenghua Liu, Yiwei Wang, Baolong Bi, Yuyao Ge, Jun Wan, Yurong Wu, Xueqi Cheng

Beyond One-Size-Fits-All: Tailored Benchmarks for Efficient Evaluation (2025.acl-long)

Challenge:	Large Language Models (LLMs) have made remarkable advances in reasoning, yet continue to struggle with hallucinations, logical errors, and inability to self-correct during complex multi-step tasks.
Approach:	They propose a framework that enhances LLM reasoning through real-time environmental feedback validating each reasoning step, dynamic branch exploration for investigating alternative solution paths when faced with errors, and experience-based learning from successful reasoning trajectories.
Outcome:	The proposed model outperforms comparable models by 24.4 percentage points across benchmarks while outperforming comparable models.

Copied to clipboard

Peiwen Yuan, Yueqi Zhang, Shaoxiong Feng, Yiwei Li, Xinglin Wang, Jiayi Shi, Chuyi Tan, Boyuan Pan, Yao Hu, Kan Li

Green & Sustainable NLP

Challenge:	Existing efficient methods estimate performance of models on large benchmarks, but these methods rely on the assumption that target models have high prediction consistency with source models.
Approach:	They propose a method that conducts customized evaluation tailored to each target model.
Outcome:	The proposed method reduces the MAE of estimates by 31.4% on benchmarks across 300 models.

SLANG: New Concept Comprehension of Large Language Models (2024.emnlp-main)

Copied to clipboard

Lingrui Mei, Shenghua Liu, Yiwei Wang, Baolong Bi, Xueqi Cheng

AlignedCoT: Prompting Large Language Models via Native-Speaking Demonstrations (2024.findings-emnlp)

Challenge:	Dynamic nature of language limits the adaptability of Large Language Models (LLMs) Traditionally, LLMs are trained on static data, which limits their adaptability .
Approach:	They propose a benchmark to integrate novel data and assess LLMs’ ability to comprehend emerging concepts, alongside a causal inference-based approach to enhance LLM comprehension of new phrases and their colloquial context.
Outcome:	The proposed model outperforms baseline models in terms of precision and relevance in the comprehension of Internet slang and memes.

Copied to clipboard

Zhicheng Yang, Yinya Huang, Jing Xiong, Liang Feng, Xiaodan Liang, Yiwei Wang, Jing Tang

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

Challenge:	Existing LLMs are delicate and elusive in prompt words and styles.
Approach:	They propose an LLM-acquainted prompting technique that includes proficient "native-speaking" they propose to use in-context learning to prompt LLMs to perform high-performance reasoning .
Outcome:	The proposed technique achieves step-wise prompts in zero-shot scenarios while maintaining the prompt quality.

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.

Calibrating Inference Time Alignment with Sequence-level Risk Accumulation (2026.acl-long)

Copied to clipboard

Shanwen Tan, Ziyang Dong, Wei Ju, Yiwei Fu, Hao Wu, Kun Wang, Yifan Wang, Ziyue Qiao

Can Graph Descriptive Order Affect Solving Graph Problems with LLMs? (2025.acl-long)

Challenge:	Existing approaches to decode large language models (LLMs) often over-reject benign information, limiting their generalizability in real-world scenarios where harmful and benign information coexist.
Approach:	They propose a framework to regulate decoding alignments for Large Language Models (LLMs) they employ a reward-guided branch decoding paradigm to incorporate safety awareness during generation.
Outcome:	The proposed framework achieves superior performance on four attack benchmarks and two neutral datasets.

Copied to clipboard

Yuyao Ge, Shenghua Liu, Baolong Bi, Yiwei Wang, Lingrui Mei, Wenjie Feng, Lizhe Chen, Xueqi Cheng

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

Challenge:	Large language models (LLMs) have achieved significant success in reasoning tasks, including mathematical reasoning and logical deduction.
Approach:	They conduct the first comprehensive analysis of how the order of graph descriptions impacts LLM performance.
Outcome:	The results show that graph descriptions significantly improve LLMs’ comprehension of graph structures, and the robustness of LLM models to graph description order varies across different tasks.

Copied to clipboard

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

DatawiseAgent: A Notebook-Centric LLM Agent Framework for Adaptive and Robust Data Science Automation (2025.emnlp-main)

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

Ziming You, Yumiao Zhang, Dexuan Xu, Yiwei Lou, Yandong Yan, Wei Wang, Huamin Zhang, Yu Huang

Robustness in NLP

Challenge:	Existing large language model (LLM) agents for data science automation are limited by narrow task scopes, limited generalization across tasks and models, and over-reliance on state-of-the-art (SOTA) LLMs.
Approach:	They propose a notebook-centric LLM agent framework for adaptive and robust data science automation.
Outcome:	The proposed framework surpasses baselines such as AutoGen and TaskWeaver in performance tests across diverse data science scenarios and models.

“Not Aligned” is Not “Malicious”: Being Careful about Hallucinations of Large Language Models’ Jailbreak (2025.coling-main)

Copied to clipboard

Lingrui Mei, Shenghua Liu, Yiwei Wang, Baolong Bi, Jiayi Mao, Xueqi Cheng

InFoBench: Evaluating Instruction Following Ability in Large Language Models (2024.findings-acl)

Challenge:	“Jailbreak” is a major safety concern of Large Language Models (LLMs).
Approach:	They propose a benchmarking framework to evaluate "jailbreak" outputs . they propose specialized validation framework to ensure outputs are useful malicious instructions .
Outcome:	The proposed framework enhances existing benchmarks to ensure outputs are useful . it also aims to evaluate the true potential of jailbroken outputs to cause harm to human society.

Copied to clipboard

Yiwei Qin, Kaiqiang Song, Yebowen Hu, Wenlin Yao, Sangwoo Cho, Xiaoyang Wang, Xuansheng Wu, Fei Liu, Pengfei Liu, Dong Yu

CogLM: Tracking Cognitive Development of Large Language Models (2025.naacl-long)

Challenge:	Existing methods for evaluating Large Language Models (LLMs) ability to follow instructions have not been able to provide a detailed analysis of their compliance with instructions.
Approach:	They propose a new metric for evaluating Large Language Models' ability to follow instructions and a benchmark for DRFR.
Outcome:	The proposed metric and benchmark compared with traditional scoring methods and explores annotation sources including human experts, crowd-sourced workers, and GPT-4.

Copied to clipboard

Xinglin Wang, Peiwen Yuan, Shaoxiong Feng, Yiwei Li, Boyuan Pan, Heda Wang, Yao Hu, Kan Li

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

Challenge:	Large Language Models (LLMs) have recently shown remarkable abilities across a wide variety of tasks, but few studies have explored the reasons behind the evolutionary relationship among various abilities.
Approach:	They construct a benchmark CogLM based on Piaget's Theory of Cognitive Development (PTC) which measures the cognitive levels of Large Language Models (LLMs) using 1,220 questions spanning 10 cognitive abilities crafted by more than 20 human experts.
Outcome:	The proposed framework provides a comprehensive testbed for the cognitive levels of LLMs.

Copied to clipboard

Cheng Wang, Yiwei Wang, Yujun Cai, Bryan Hooi

How to Make Large Language Models Generate 100% Valid Molecules? (2025.emnlp-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

Wen Tao, Jing Tang, Alvin Chan, Bryan Hooi, Baolong Bi, Nanyun Peng, Yuansheng Liu, Yiwei Wang

Vulnerability of Text-to-Image Models to Prompt Template Stealing: A Differential Evolution Approach (2025.findings-acl)

Challenge:	Large language models (LLMs) can learn to perform a wide range of tasks, but generating valid molecules using representations like SMILES is challenging in few-shot settings.
Approach:	They propose a language framework that converts invalid SMILES to SELFIES and LLMs as post-hoc correctors to ensure that the molecules generated by LLM are 100% valid.
Outcome:	The proposed model performs worse with SELFIES than with SMILES and improves on other metrics.

Copied to clipboard

Yurong Wu, Fangwen Mu, Qiuhong Zhang, Jinjing Zhao, Xinrun Xu, Lingrui Mei, Yang Wu, Lin Shi, Junjie Wang, Zhiming Ding, Yiwei Wang

Mitigating Lost-in-Retrieval Problems in Retrieval Augmented Multi-Hop Question Answering (2025.acl-long)

Challenge:	Prompt trading has emerged as a significant intellectual property concern in recent years, where vendors entice users by showcasing sample images before selling prompt templates that can generate similar images.
Approach:	They propose a prompt-stealing benchmark consisting of 50 templates and 450 images organized into Easy and Hard difficulty levels.
Outcome:	The proposed method outperforms baseline methods with an average improvement of over 10%.

Copied to clipboard

Rongzhi Zhu, Xiangyu Liu, Zequn Sun, Yiwei Wang, Wei Hu

Natural Language Interfaces

Challenge:	Empirical results show that ChainRAG consistently outperforms baselines in both effectiveness and efficiency.
Approach:	They propose a method which sequentially handles each sub-question by completing missing key entities and retrieving relevant sentences from a sentence graph for answer generation.
Outcome:	The proposed method outperforms baselines on three multi-hop QA datasets.

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

Copied to clipboard

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

Speculative Decoding for Multi-Sample Inference (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

Yiwei Li, Jiayi Shi, Shaoxiong Feng, Peiwen Yuan, Xinglin Wang, Yueqi Zhang, Ji Zhang, Chuyi Tan, Boyuan Pan, Yao Hu, Kan Li

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

Challenge:	Speculative decoding method exploits consensus of parallel reasoning paths to synthesize high-quality draft tokens without auxiliary models or external databases.
Approach:	They propose a speculative decoding method that exploits the consensus of parallel reasoning paths to synthesize high-quality draft tokens without auxiliary models or external databases.
Outcome:	The proposed method exploits the intrinsic consensus of parallel reasoning paths to synthesize high-quality draft tokens without auxiliary models or databases.

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

Adaptive Token Biaser: Knowledge Editing via Biasing Key Entities (2024.findings-emnlp)

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

Baolong Bi, Shenghua Liu, Yiwei Wang, Lingrui Mei, Hongcheng Gao, Yilong Xu, Xueqi Cheng

Challenge:	Existing methods to update parametric knowledge of large language models (LLMs) are outdated and incontext editing (KE) is not effective due to the substantial cost associated with retraining.
Approach:	They propose a new decoding technique that enhances in-context editing (ICE) they propose to use parametric knowledge to update the models' knowledge .
Outcome:	The proposed technique improves ICE performance while incurring only half the latency.

From Logical to Computational Sparsity: Structure-Aware Block-Sparse Attention for Long-Code Completion (2026.acl-long)

Copied to clipboard

Yanli Wang, Yanlin Wang, Bowen Zhang, Yiwei Zhang, Daya Guo, Jiachi Chen, Hongyu Zhang, Zibin Zheng

Programming Languages in NLP

Challenge:	Existing sparse attention methods for long-context generation pose high latency . general sparsity methods cause excessive accuracy degradation without considering code structure .
Approach:	They propose a training-free Structure-aware block-sparse attention mechanism that bridges the gap between logical and computational sparsity.
Outcome:	The proposed method reduces TTFT by 45-55% while maintaining accuracy within 3% of dense attention.

SQLAgent: Learning to Explore Before Generating as a Data Engineer (2026.findings-acl)

Copied to clipboard

Wenjia Jiang, Yiwei Wang, Boyan Han, Joey Tianyi Zhou, Chi Zhang

Programming Languages in NLP

Challenge:	Existing approaches to large language models fail to generalize in complex, real-world settings due to database-specific nature of SQL reasoning.
Approach:	They propose a two-stage LLM-based framework that decouples knowledge acquisition from query generation.
Outcome:	The proposed framework significantly improves accuracy over baselines on large-scale benchmarks.

METAL: A Multi-Agent Framework for Chart Generation with Test-Time Scaling (2025.acl-long)

Copied to clipboard

Bingxuan Li, Yiwei Wang, Jiuxiang Gu, Kai-Wei Chang, Nanyun Peng

Who is in the Spotlight: The Hidden Bias Undermining Multimodal Retrieval-Augmented Generation (2025.emnlp-main)

Challenge:	Chart generation requires strong visual design skills and precise coding capabilities that embed the desired visual properties into code.
Approach:	They propose a vision-language model-based multi-agent framework for effective automatic chart generation.
Outcome:	The proposed framework achieves a 5.2% improvement in the F1 score over the current best chart generation task.

Copied to clipboard

Jiayu Yao, Shenghua Liu, Yiwei Wang, Lingrui Mei, Baolong Bi, Yuyao Ge, Zhecheng Li, Xueqi Cheng

Information Retrieval

Challenge:	Existing RAG models are sensitive to the order in which evidence is presented, resulting in unstable performance and biased reasoning.
Approach:	They propose to quantify position bias in multimodal RAG systems by using position sensitivity index . they also develop a visualization framework to trace attention allocation patterns across decoder layers .
Outcome:	The proposed framework shows that multimodal interactions intensify position bias compared to unimodal settings and that this bias increases logarithmically with retrieval range.

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

Copied to clipboard

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

From Sub-Ability Diagnosis to Human-Aligned Generation: Bridging the Gap for Text Length Control via MarkerGen (2025.acl-long)

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

Peiwen Yuan, Chuyi Tan, Shaoxiong Feng, Yiwei Li, Xinglin Wang, Yueqi Zhang, Jiayi Shi, Boyuan Pan, Yao Hu, Kan Li

Challenge:	Existing methods to control text length are lacking in LCTG, posing a major limitation for practical applications.
Approach:	They propose a plug-and-play approach that decomposes LCTG sub-abilities with human patterns as reference and performs detailed error analysis.
Outcome:	The proposed method significantly improves LCTG across various settings, exhibiting outstanding effectiveness and generalizability.

Hard to Read, Easy to Jailbreak: How Visual Degradation Bypasses MLLM Safety Alignment (2026.findings-acl)

Copied to clipboard

Zhixue Song, Boyan Han, Yiwei Wang, Chi Zhang

Revisiting Self-Consistency from Dynamic Distributional Alignment Perspective on Answer Aggregation (2025.findings-acl)

Challenge:	Recent advances in visual context compression enable MLLMs to process ultra-long contexts efficiently by rendering text into images.
Approach:	They propose a strategy that decouples visual transcription from safety auditing by enforcing a serialized pipeline to decouple visual transcription and safety assessment.
Outcome:	The proposed strategy decouples visual transcription from safety auditing to reduce the risk of jailbreaking.

Copied to clipboard

Yiwei Li, Ji Zhang, Shaoxiong Feng, Peiwen Yuan, Xinglin Wang, Jiayi Shi, Yueqi Zhang, Chuyi Tan, Boyuan Pan, Yao Hu, Kan Li

Embodied-Reasoner: Synergizing Visual Search, Reasoning, and Action for Embodied Interactive Tasks (2026.acl-long)

Challenge:	Existing studies on self-consistency show that it improves reasoning abilities by aggregating diverse stochastic samples.
Approach:	They propose a confidence-driven mechanism that dynamically calibrates temperature to align with high probability modes.
Outcome:	The proposed method outperforms fixed-diversity baselines on reasoning tasks and improves both average and best-case performance.

Copied to clipboard

Wenqi Zhang, Mengna Wang, Gangao Liu, Huixin Xu, Yiwei Jiang, Yongliang Shen, Guiyang Hou, Zhe Zheng, Hang Zhang, Xin Li, Jiajun Liu, Weiming Lu, Peng Li, Yueting Zhuang

MoEC: A Memory-Routed Mixture-of-Experts Controller for Adaptive Minecraft Control (2026.acl-long)

Challenge:	Recent advances in reasoning models have demonstrated remarkable capabilities on mathematical and coding tasks, but their effectiveness in embodied domains remains largely unexplored.
Approach:	They propose a reasoning model for interactive embodied tasks that synthesizes 9.3k coherent Observation-Thought-Action trajectories containing 64k ego-centric images and 90k diverse reasoning processes.
Outcome:	The proposed model outperforms existing visual reasoning models by +9%, 24%, and +13% on long-horizon tasks.

Copied to clipboard

Hui Wu, Chao Xu, Jianghui Wang, Ziqiong Liu, Dong Li, Yiwei Dai, Emad Barsoum

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

Challenge:	Existing systems rely on a monolithic policy to execute subgoals across varying contexts, causing inconsistent outcomes and scaling only partially mitigates.
Approach:	They propose a memory-routed mixtureof-experts controller for Adaptive Minecraft Control that routes via a subgoal-indexed expert memory and regulates capacity through failure-triggered expert growth and redundancy-aware consolidation.
Outcome:	The proposed controller shows significant gains in adaptability, robustness, and execution consistency over strong baselines.

Copied to clipboard

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

HiddenGuard: Fine-Grained Safe Generation with Specialized Representation Router (2026.acl-long)

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

Lingrui Mei, Shenghua Liu, Yiwei Wang, Baolong Bi, Ruibin Yuan, Xueqi Cheng

TeCES: Collaborative Geometric Knowledge Representation Framework under Evolving Fact Snapshots (2026.acl-long)

Challenge:	Current alignment approaches rely on refusal alignment to avoid harmful content . large language models are often overly cautious or overlook subtle harmful content.
Approach:	They propose a framework for fine-grained safe generation in Large Language Models that enables real-time, token-level harmfulness detection and redaction without loss in capability.
Outcome:	The proposed framework achieves over 90% in F1 score for detecting and redacting harmful content while preserving overall utility and informativeness of the model’s responses.

Copied to clipboard

Jiujiang Guo, Zhengliang Guo, Kai Wang, Meiyang Wang, Dehua Peng, Shaozu Yuan, Chengyin Hu, Shuan Ai, Yiwei Wei

Vulnerability of Large Language Models to Output Prefix Jailbreaks: Impact of Positions on Safety (2025.findings-naacl)

Challenge:	Existing knowledge graphs represent static facts but lack collaborative modeling of both . e.g., existing knowledge graph models lack a framework for integrating snapshots into knowledge graph.
Approach:	They propose a framework for high-fidelity modeling of evolving snapshots using concept of snapshots.
Outcome:	The proposed framework outperforms existing models on six benchmarks.

Copied to clipboard

Yiwei Wang, Muhao Chen, Nanyun Peng, Kai-Wei Chang

Generative Dense Retrieval: Memory Can Be a Burden (2024.eacl-long)

Challenge:	Previous research on jailbreak attacks has focused on optimizing the adversarial snippet content injected into input prompts to expose LLM security vulnerabilities.
Approach:	They propose to use a simple adversarial snippet at the beginning of output to expose LLM security vulnerabilities.
Outcome:	The proposed approach exposes LLM security vulnerabilities much faster than input suffix attacks or prompt-based output jailbreaks.

Copied to clipboard

Peiwen Yuan, Xinglin Wang, Shaoxiong Feng, Boyuan Pan, Yiwei Li, Heda Wang, Xupeng Miao, Kan Li

Text Clustering

Challenge:	Empirical results show that Generative Dense Retrieval (GDR) achieves an average of 3.0 R@100 improvement on NQ dataset under multiple settings and has better scalability.
Approach:	They propose a Generative Dense Retrieval paradigm that auto-decodes document identifiers given a query and uses memory to avoid memory confusion.
Outcome:	Empirical results show that the proposed paradigm improves on the small-scale corpora and improves scalability.

Harmonizing Dense and Sparse Signals in Multi-turn RL: Dual-Horizon Credit Assignment for Industrial Sales Agents (2026.acl-industry)

Copied to clipboard

Haojin Yang, Ai Jian, Yiwei Wang, Xinyue Huang, Weipeng Zhang, Ke Zeng, Xunliang Cai, Jingqing Ruan

Challenge:	Large language models for industrial sales require balancing long-term commercial objectives with immediate linguistic constraints such as fluency and compliance.
Approach:	They propose a framework that disentangles optimization across time scales by normalizing advantages from turn-level and session-level rewards before fusion.
Outcome:	The proposed framework outperforms the state-of-the-art GRPO model in conversion rate and identity detection rate.

Poor-Supervised Evaluation for SuperLLM via Mutual Consistency (2024.findings-acl)

Copied to clipboard

Peiwen Yuan, Shaoxiong Feng, Yiwei Li, Xinglin Wang, Boyuan Pan, Heda Wang, Yao Hu, Kan Li