Papers by Yuchen Wu
Reason-KE++: Aligning the Process, Not Just the Outcome, for Faithful LLM Knowledge Editing (2026.findings-acl)
Copied to clipboard
| Challenge: | Current methods for modifying parameters to integrate new knowledge are not accurate enough. |
| Approach: | They propose an SFT+RL framework that instills process-level faithfulness by a stage-aware Reward mechanism and a Stage-assisted Reward Mechanism. |
| Outcome: | The proposed framework instills process-level faithfulness while boosting final accuracy. |
CARO: Chain-of-Analogy Reasoning Optimization for Robust Content Moderation (2026.findings-acl)
Copied to clipboard
| Challenge: | Current large language models struggle with ambiguous content moderation cases due to misleading "decision shortcuts" . authors propose a two-stage training framework to induce robust analogical reasoning in LLMs . |
| Approach: | They propose a two-stage training framework to induce robust analogical reasoning in LLMs . they bootstrap analogy reasoning chains via retrieval-augmented generation and SFT . |
| Outcome: | The proposed framework outperforms state-of-the-art reasoning models and specialized moderation models on ambiguous moderation benchmarks. |
Uncertainty-Aware Routing for Principled Alignment with MoE Dynamics (2026.acl-long)
Copied to clipboard
Yilong Chen, Junyuan Shang, Yuchen Feng, Zhenyu Zhang, Naibin Gu, Ziqi Wang, Tingwen Liu, Shuohuan Wang, Yu Sun, Hua Wu, Haifeng Wang
| Challenge: | Mixture-of-Experts (MoE) is a cornerstone for scaling LLMs, yet its training dynamics remain poorly understood, often leading to sub-optimal specialization. |
| Approach: | They propose to use Helmholtz Free Energy and Router Entropy to study the MoE lifecycle and identify a universal Three-Stage Phase Transition . |
| Outcome: | The proposed model reduces perplexity and improves expert distinctiveness, offering a principled path toward thermodynamically aligned computation. |
MedAdapter: Efficient Test-Time Adaptation of Large Language Models Towards Medical Reasoning (2024.emnlp-main)
Copied to clipboard
| Challenge: | Large language models (LLMs) have improved generation and reasoning capabilities compared to traditional BERT-sized models due to massive number of parameters and extensive pre-training on vast textual corpora. |
| Approach: | They propose a unified post-hoc adapter for test-time adaptation of large language models . they propose to fine-tune only a small BERT-sized adapter to rank candidate LLMs . |
| Outcome: | The proposed adapter improves performance on four biomedical tasks without requiring computational resources or sharing data with third parties. |
Rethinking Text-to-SQL: Dynamic Multi-turn SQL Interaction for Real-world Database Exploration (2026.findings-acl)
Copied to clipboard
Linzhuang Sun, Tianyu Guo, Hao Liang, Ruitong Liu, Yuying Li, Qifeng Cai, Jingxuan Wei, Yuchen Wu, Bihui Yu, Xiangxiang Zhang, Wentao Zhang, Bin Cui
| Challenge: | Structured Query Language (SQL) is the cornerstone for data-driven decision-making. |
| Approach: | They propose a benchmark to rigorously evaluate Large Language Models within a dynamic interaction framework. |
| Outcome: | The proposed benchmark aims to rigorously evaluate LLMs within a dynamic interaction framework. |
MMBoundary: Advancing MLLM Knowledge Boundary Awareness through Reasoning Step Confidence Calibration (2025.acl-long)
Copied to clipboard
| Challenge: | Existing methods calibrate model confidence on entire response, which leads to incorrect answers with high confidence. |
| Approach: | They propose a framework that advances the knowledge boundary awareness of multimodal large language models through reasoning step confidence calibration. |
| Outcome: | Empirical results show that the proposed framework outperforms existing methods across domains and metrics. |
OptiVerse: A Comprehensive Benchmark towards Optimization Problem Solving (2026.findings-acl)
Copied to clipboard
| Challenge: | Existing benchmarks focus on Mathematical Programming and Combinatorial Optimization, hindering comprehensive evaluation. |
| Approach: | They propose a benchmarking tool that compares 1,000 curated optimization problems across three difficulty levels. |
| Outcome: | The proposed model improves performance on hard problems while maintaining 27% accuracy. |
EHRAgent: Code Empowers Large Language Models for Few-shot Complex Tabular Reasoning on Electronic Health Records (2024.emnlp-main)
Copied to clipboard
Wenqi Shi, Ran Xu, Yuchen Zhuang, Yue Yu, Jieyu Zhang, Hang Wu, Yuanda Zhu, Joyce Ho, Carl Yang, May Dongmei Wang
| Challenge: | EHRAgent enables clinicians to interact with EHRs using natural language . reliance on rule-based conversion systems often necessitates additional training or effort from data engineers. |
| Approach: | They propose a large language model agent that generates and executes code in natural language to facilitate clinicians in directly interacting with EHRs. |
| Outcome: | The proposed agent outperforms the strongest baseline by up to 29.6% in success rate on three real-world EHR datasets. |
On Fake News Detection with LLM Enhanced Semantics Mining (2024.emnlp-main)
Copied to clipboard
| Challenge: | Existing methods for detecting fake news use only news embeddings to capture the lexical semantics between tokens. |
| Approach: | They propose a topic-based model with prompts to extract news embeddings from LLMs and a generalized page-rank model to extract local and global semantics. |
| Outcome: | The proposed model shows superior performance on five benchmark datasets over seven baseline methods. |
Pause or Fabricate? Training Language Models for Grounded Reasoning (2026.findings-acl)
Copied to clipboard
Yiwen Qiu, Linjuan Wu, Yizhou Liu, Yuchen Yan, Jin Ma, Xu Tan, Yao Hu, Daoxin Zhang, Wenqi Zhang, Weiming Lu, Jun Xiao, Yongliang Shen
| Challenge: | Large language models implicitly fabricate information when inputs are incomplete, causing confidence but unreliable conclusions. |
| Approach: | They propose a framework for grounded reasoning under incomplete information that decomposes reasoning into two stages . they propose stage-specific rewards to penalize hallucinations, enabling models to detect gaps, stop proactively, and resume reasoning after clarification. |
| Outcome: | The proposed framework improves premise detection and task success by 30% . it also reduces average response length by over 20% . |
Adversarial Yet Cooperative: Multi-Perspective Reasoning in Retrieved-Augmented Language Models (2026.findings-acl)
Copied to clipboard
Can Xu, Lingyong Yan, Jiayi Wu, Haosen Wang, Shuaiqiang Wang, Yuchen Li, Jizhou Huang, Dawei Yin, Xiang Li
| Challenge: | Existing training paradigms rely on outcome-oriented rewards, which provide insufficient signal for shaping the complex, multi-step reasoning process. |
| Approach: | They propose a framework that integrates large reasoning models with retrieval-augmented generation to improve reasoning fidelity and verification rigor. |
| Outcome: | Experiments on multiple benchmarks demonstrate the effectiveness of the proposed framework. |
Label-Driven Denoising Framework for Multi-Label Few-Shot Aspect Category Detection (2022.findings-emnlp)
Copied to clipboard
| Challenge: | Existing methods for ACD use label information of aspect categories to detect aspect categories . but, they still suffer from noise problems due to lack of supervised data . |
| Approach: | They propose a Label-Driven Denoising Framework to alleviate noise problems for ACD subtask . they use the label information of each aspect to generate a better prototype . |
| Outcome: | The proposed framework improves the performance of the multi-label few-shot Aspect Category Detection task. |
AskToAct: Enhancing LLMs Tool Use via Self-Correcting Clarification (2025.emnlp-main)
Copied to clipboard
Xuan Zhang, Yongliang Shen, Zhe Zheng, Linjuan Wu, Wenqi Zhang, Yuchen Yan, Qiuying Peng, Jun Wang, Weiming Lu
| Challenge: | Existing tools for ambiguous and incomplete queries are limited by manual construction and lack of error correction mechanisms during multi-turn clarification. |
| Approach: | They propose a framework that exploits the mapping between queries and their tool invocation solutions by removing key parameters from queries while retaining them as ground truth. |
| Outcome: | The proposed framework outperforms existing methods while maintaining high accuracy in tool invocation. |
Rethinking Masked Language Modeling for Chinese Spelling Correction (2023.acl-long)
Copied to clipboard
| Challenge: | Existing CSC models over-fit the error model while under-fitting the language model, resulting in poor generalization to out-of-distribution error patterns. |
| Approach: | They propose to use a multi-domain benchmark LEMON to assess the open domain generalization of Chinese Spelling Correction models. |
| Outcome: | The proposed method achieves state-of-the-art results on SIGHAN, ECSpell, and LEMON. |
F-MALLOC: Feed-forward Memory Allocation for Continual Learning in Neural Machine Translation (2024.naacl-long)
Copied to clipboard
| Challenge: | Existing approaches to address Catastrophic Forgetting (CF) have been developed to avoid forgetting and maintain system extensibility. |
| Approach: | They propose a method to reduce Catastrophic Forgetting (CF) by decomposing feed-forward layers into discrete memory cells and ensuring robust extendability. |
| Outcome: | The proposed method achieves higher BLEU scores and almost zero forgetting while maintaining robust extendability. |
Cultivating Forensic Reasoning for Generalizable Multimodal Manipulation Detection (2026.acl-long)
Copied to clipboard
| Challenge: | Existing methods for manipulation detection and grounding focus on manipulator type classification under result-oriented supervision. |
| Approach: | They propose a reasoning-driven framework that shifts learning from outcome fitting to process modeling. |
| Outcome: | The proposed framework achieves state-of-the-art with superior generalization on large-scale datasets. |
RAP-ID: Mechanistic Prompt Injection Detection via Impostor Behavior Analysis (2026.findings-acl)
Copied to clipboard
| Challenge: | Existing defenses rely on externally deployed guardrail models or response inspection . current defenses depend on external guardrails or response inspecting . |
| Approach: | They propose a mechanistic, train-free detection framework that operates exclusively on internal state dynamics during the initial forward pass. |
| Outcome: | The proposed framework achieves competitive performance with significant overall improvements compared to heuristic methods. |
Dynamic Generation of Multi LLM Agents Communication Topologies with Graph Diffusion Models (2026.acl-long)
Copied to clipboard
Eric Hanchen Jiang, Levina Li, Frank Wan, Xiao Liang, Sophia Yin, Yuchen Wu, Xinfeng Li, Yizhou Sun, Wei Wang, Kai-Wei Chang, Ying Nian Wu
| Challenge: | Existing frameworks rely on static or rule-based topologies that fail to adapt to task requirements. |
| Approach: | They propose a generative framework that generates highly task-adaptive topologies . they validated the framework on multiple benchmarks and validated it on multiple platforms . |
| Outcome: | The proposed framework outperforms existing frameworks in task-adaptive communication topologies. |
Robust Knowledge Editing via Explicit Reasoning Chains for Distractor-Resilient Multi-Hop QA (2025.findings-emnlp)
Copied to clipboard
| Challenge: | Large language models encode vast amounts of knowledge but remain static once trained, making timely integration of emerging facts prohibitively expensive via full retraining. |
| Approach: | They introduce a reasoning-chain-based editing framework that steers a pretrained LLM through four structured stages to filter distractors in a single pass. |
| Outcome: | The proposed framework steers a pretrained LLM through four structured stages to filter distractors in a single pass. |
CriticLean: Critic-Guided Reinforcement Learning for Mathematical Formalization (2026.acl-long)
Copied to clipboard
Zhongyuan Peng, Yifan Yao, Kaijing Ma, Shuyue Guo, Yizhe Li, Yichi Zhang, Chenchen Zhang, Yifan Zhang, Zhouliang Yu, Luming Li, Minghao Liu, Yihang Xia, Jiawei Shen, Yuchen Wu, Yixin Cao, Zhaoxiang Zhang, Wenhao Huang, Jiaheng Liu, Ge Zhang
| Challenge: | Existing approaches to formalizing mathematical statements face limitations in accuracy, especially in the context of complex, highlevel problems that involve sophisticated mathematical reasoning. |
| Approach: | They propose a CriticLean framework that elevates the role of the critic from a passive validator to an active learning component and introduce a benchmark to measure models’ ability to distinguish semantically correct from incorrect formalizations. |
| Outcome: | The proposed framework outperforms open- and closed-source benchmarks and shows that it significantly outperformed existing models. |
Edit Once, Update Everywhere: A Simple Framework for Cross-Lingual Knowledge Synchronization in LLMs (2025.findings-acl)
Copied to clipboard
| Challenge: | Existing methods to update large language models focus on single-language editing or basic multilingual editing, failing to achieve true cross-linguistic knowledge synchronization. |
| Approach: | They propose a cross-linguistic knowledge democracy edit technique to improve cross-lingual performance. |
| Outcome: | The proposed method improves cross-lingual performance while maintaining high accuracy in monolingual settings. |
CHAIRO: Contextual Hierarchical Analogical Induction and Reasoning Optimization for LLMs (2026.acl-long)
Copied to clipboard
| Challenge: | Recent advances in large language models (LLMs) have enabled more sophisticated content moderation, but these methods lack generalization, interpretability, and adaptability to unseen or ambiguous cases. |
| Approach: | They propose a new moderation framework that leverages analogical examples to enhance rule induction and decision reliability. |
| Outcome: | The proposed method outperforms rule-injected fine-tuning baselines and multi-stage static RAG pipelines in terms of moderation accuracy and rule quality. |
A Survey of Reinforcement Learning for Large Language Models under Data Scarcity: Challenges and Solutions (2026.acl-long)
Copied to clipboard
Zhiyin Yu, Yuchen Mou, Juncheng Yan, Junyu Luo, Chunchun Chen, Xing Wei, Yunhui Liu, Hongru Sun, Yuxing Zhang, Jun Xu, Yatao Bian, Ming Zhang, Wei Ye, Tieke He, Jie Yang, Guanjie Zheng, Zhonghai Wu, Bo Zhang, Lei Bai, Xiao Luo
| Challenge: | Existing research on reinforcement learning for LLMs under data scarcity has not been unified. |
| Approach: | They propose a top-up hierarchical framework built around three complementary perspectives: data-centric, training-centric and framework-centric. |
| Outcome: | The proposed framework provides a clear conceptual foundation for understanding the design space of data-efficient RL for large language models and to guide researchers working in this emerging area. |
A Thorough Examination on Zero-shot Dense Retrieval (2023.findings-emnlp)
Copied to clipboard
Ruiyang Ren, Yingqi Qu, Jing Liu, Xin Zhao, Qifei Wu, Yuchen Ding, Hua Wu, Haifeng Wang, Ji-Rong Wen
| Challenge: | Recent advances in dense retrieval (DR) models have been shown to be not as competitive as traditional sparse retrieval models in a zero-shot retrieval setting. |
| Approach: | They propose to examine the zero-shot capability of DR models by analyzing key factors related to source training set and potential bias from target dataset. |
| Outcome: | The proposed model is not as competitive as sparse retrieval models in a zero-shot retrieval setting. |
RocketQA: An Optimized Training Approach to Dense Passage Retrieval for Open-Domain Question Answering (2021.naacl-main)
Copied to clipboard
Yingqi Qu, Yuchen Ding, Jing Liu, Kai Liu, Ruiyang Ren, Wayne Xin Zhao, Daxiang Dong, Hua Wu, Haifeng Wang
| Challenge: | Open-domain question answering uses dense passage retrieval to find answers . however, it is difficult to effectively train a dual-encoder due to discrepancy between training and inference . |
| Approach: | They propose an optimized training approach to improve dense passage retrieval using RocketQA . they propose cross-batch negatives, denoised hard negatives and data augmentation . |
| Outcome: | The proposed approach outperforms state-of-the-art models on both MSMARCO and Natural Questions. |