Papers by Yuhan Chen

The Stepwise Deception: Simulating the Evolution from True News to Fake News with LLM Agents (2025.emnlp-main)

Challenge:	Existing methods to induce Chain-of-Thought (CoT) in LLMs are limited and do not consider the importance of efficiently utilizing existing CoT data.
Approach:	They propose a new training paradigm which exploits the inherent information in CoT for iterative generation.
Outcome:	The proposed training paradigm surpasses direct seq2seq training on CoT-extensive tasks without data augmentation or altering the model itself.

Copied to clipboard

Yuhan Liu, Zirui Song, Juntian Zhang, Xiaoqing Zhang, Xiuying Chen, Rui Yan

Enhancing Attributed Question Answering using Tailored Progressive Curriculum Learning (2025.findings-emnlp)

Challenge:	Existing studies assume fake news is inherently existing rather than exploring its gradual formation.
Approach:	They propose a Large Language Model-based simulation approach explicitly focusing on fake news evolution from real news.
Outcome:	The proposed framework captures fake news evolution patterns and accurately reproduces known fake news, aligning closely with human evaluations.

Copied to clipboard

Yuhan Chen, Bowei Zou, Yifan Fan, Yuchong Chen, Shujun Cao, Yu Hong

Question Answering

Challenge:	Attributed Question Answering models are not yet leveraged to enhance their essential capabilities, including evidence identification, cross-source relation recognition and anti-distraction reasoning.
Approach:	They propose a progressive progressive curriculum learning approach that optimizes both encoder-decoder and decoder-only AQA models.
Outcome:	The proposed approach improves both encoder-decoder and decoder-only AQA models on the quotesum benchmark.

SDiaReward: Modeling and Benchmarking Spoken Dialogue Rewards with Modality and Colloquialness (2026.acl-long)

Copied to clipboard

Jingyu Lu, Yuhan Wang, Fan Zhuo, Xize Cheng, Changhao Pan, Xueyi Pu, Yifu Chen, Chenyuhao Wen, Tianle Liang, Zhou Zhao

Speech & Audio in NLP

Challenge:	SDiaReward is an end-to-end spoken dialogue system that integrates paralinguistic nuances and spontaneous nature of human conversation.
Approach:	They propose an end-to-end multi-turn reward model trained on SDiaReward-Dataset . it is a collection of episode-level preference pairs targeting modality and colloquiality gaps .
Outcome:	The proposed model outperforms general-purpose audio LLMs in episode-level evaluation.

DialoGPS: Dialogue Path Sampling in Continuous Semantic Space for Data Augmentation in Multi-Turn Conversations (2023.acl-long)

Copied to clipboard

Ang Lv, Jinpeng Li, Yuhan Chen, Gao Xing, Ji Zhang, Rui Yan

Dialogue Systems & Conversational Agents

Challenge:	a context leads to various responses, and a response answers multiple contexts.
Approach:	They propose a method that augments open-domain dialogue generation from a many-to-many perspective.
Outcome:	The proposed method can augment open-domain dialogue generation tasks with automatic and human evaluation.

Injecting Domain-Specific Knowledge into Large Language Models: A Comprehensive Survey (2025.findings-emnlp)

Copied to clipboard

Zirui Song, Bin Yan, Yuhan Liu, Miao Fang, Mingzhe Li, Rui Yan, Xiuying Chen

FlexiQA: Leveraging LLM’s Evaluation Capabilities for Flexible Knowledge Selection in Open-domain Question Answering (2024.findings-eacl)

Challenge:	specialized LLMs are often limited in domain-specific applications that require specialized knowledge.
Approach:	They provide a comprehensive overview of four key methods to enhance large language models by integrating domain-specific knowledge.
Outcome:	The proposed methods are categorized into four key approaches: dynamic knowledge injection, static knowledge embedding, modular adapters, and prompt optimization.

Copied to clipboard

Yuhan Chen, Shuqi Li, Rui Yan

FinSafetyBench: Evaluating LLM Safety in Real-World Financial Scenarios (2026.findings-acl)

Challenge:	Current methods for open-domain question answering lacks the hallucination and relevance of acquired knowledge to the given question.
Approach:	They propose a new pipeline that utilizes the diverse evaluation capabilities of large language models to select knowledge effectively and flexibly.
Outcome:	The proposed pipeline combines the strengths of both paradigms and overcomes their shortcomings.

Copied to clipboard

Yutao Hou, Yihan Jiang, Yuhan Xie, Jian Yang, Liwen Zhang, Hailiang Huang, Guanhua Chen, Yun Chen

Fortify the Shortest Stave in Attention: Enhancing Context Awareness of Large Language Models for Effective Tool Use (2024.acl-long)

Challenge:	Existing large language models (LLMs) are prone to misuse and misinformation, posing serious compliance risks.
Approach:	They propose a bilingual red-teaming benchmark to test an LLM’s refusal of requests that violate financial compliance.
Outcome:	The proposed benchmark is based on real-world financial crime cases and ethical violations and includes 14 subcategories covering financial crimes and ethical breaches.

Copied to clipboard

Yuhan Chen, Ang Lv, Ting-En Lin, Changyu Chen, Yuchuan Wu, Fei Huang, Yongbin Li, Rui Yan

DeMAC: Enhancing Multi-Agent Coordination with Dynamic DAG and Manager-Player Feedback (2025.findings-emnlp)

Challenge:	In this paper, we demonstrate that an inherent waveform pattern in the attention allocation of large language models significantly affects their performance in tasks demanding a high degree of context awareness.
Approach:	They propose a method that compensates an attention trough with an attention peak by a process to enhance the model's awareness to various contextual positions.
Outcome:	The proposed method improves the performance of a 7B model on the largest tool-use benchmark, comparable to that of GPT-4.

Copied to clipboard

Yuhan Liu, Cong Xu, Lu Liu, Yihua Wang, Feiyu Chen, Qi Jia, Yaqian Zhao, Zhichun Wang, Xiang Li

Bootstrapping Code Translation with Weighted Multilanguage Exploration (2026.acl-long)

Challenge:	Multi-agent systems (MAS) powered by large language models struggle to adapt to evolving task dependencies and to handle uncertainties.
Approach:	They propose a Dynamic Environment-Aware Manager-Player Agents Coordination framework that enhances multi-agent coordination through long-term strategic planning.
Outcome:	The proposed framework outperforms traditional reinforcement learning and human-agent collaboration in the Overcooked simulation.

Copied to clipboard

Yuhan Wu, Huan Zhang, Wei Cheng, Chen Shen, Jingyue Yang, Wei Hu

Multilinguality

Challenge:	Existing methods to improve code translation depend on abundant parallel code of high quality, which may not always be available.
Approach:	They propose a method that leverages functional invariance and cross-lingual portability of test suites to serve as universal verification oracles for multilingual reinforcement learning.
Outcome:	The proposed method leverages functional invariance and cross-lingual portability of test suites to serve as universal verification oracles for multilingual reinforcement learning (RL) training.

Revisiting LoRA through the Lens of Parameter Redundancy: Spectral Encoding Helps (2025.findings-acl)

Copied to clipboard

Jiashun Cheng, Aochuan Chen, Nuo Chen, Ziqi Gao, Yuhan Li, Jia Li, Fugee Tsung

Reasoning

Challenge:	Low-Rank Adaptation (LoRA) has emerged as a prominent technique for fine-tuning large foundation models.
Approach:	They propose a low-rank Adaptation technique that harnesses the expressiveness of spectral bases to re-parameterize LoRA from a sparse spectral subspace.
Outcome:	The proposed technique achieves greater efficiency with fewer parameters than baselines on various downstream tasks, including commonsense reasoning, math reasoning, and code generation.

STEP: Success-Rate-Aware Trajectory-Efficient Policy Optimization (2026.findings-acl)

Copied to clipboard

Yuhan Chen, Yuxuan Liu, Long Zhang, Pengzhi Gao, Jian Luan, Wei Liu

Green & Sustainable NLP

Challenge:	Existing GRPO-based methods allocate sampling uniformly across tasks regardless of difficulty, propagate misleading learning signals and incur high sample-collection costs.
Approach:	They propose a framework that allocates sampling based on per-task success rates and performs fine-grained step-level optimization.
Outcome:	The proposed method improves sample efficiency and training stability over existing GRPO variants and three ablation variants on OSWorld and AndroidWorld.

Detecting AI-Generated Video: A Vision–Language Dual-View Survey (2026.findings-acl)

Copied to clipboard

Dylan Xinming Hou, Juntian Zhang, Xu Gu, Yichen Wu, Nils Lukas, Gus Xia, Xiuying Chen, Yuhan Liu

Multimodality

Challenge:	realism of AI-generated Videos (AIGC-V) rendering artifact-centric detection insufficient, authors argue . a vision–language dual-view taxonomy is proposed to systematize this rapidly evolving field .
Approach:	They propose a Vision–Language Dual-View taxonomy to systematize AIGC-V detection . they propose realism of AI-generated Videos is rendering traditional inspection insufficient .
Outcome:	The proposed model aims to show that the existing methods are consistent with real-world facts.

An Analysis and Mitigation of the Reversal Curse (2024.emnlp-main)

Copied to clipboard

Ang Lv, Kaiyi Zhang, Shufang Xie, Quan Tu, Yuhan Chen, Ji-Rong Wen, Rui Yan

Thinking Before Running! Efficient Code Generation with Thorough Exploration and Optimal Refinement (2025.findings-acl)

Challenge:	Recent research observes a phenomenon in large language models called the "reversal curse" when dealing with two entities, LLMs excel in handling sequences in the form of "aRb" but when asked "who is Mary Lee Pfeiffer's son?" the LLM exhibits considerable confusion and fails to provide a as the answer .
Approach:	They conduct the first-ever study of how the reversal curse happens in large language models . they find that LLMs excel in handling sequences in the form of "aRb" but struggle to provide a satisfactory answer when asked "who is Mary Lee Pfeiffer's son?"
Outcome:	The proposed study shows that the reversal curse can stem from specific training objectives . the study also shows that a reverse query can be difficult to understand .

Copied to clipboard

Xiaoqing Zhang, Yuhan Liu, Flood Sung, Xiuying Chen, Shuo Shang, Rui Yan

Programming Languages in NLP

Challenge:	Recent research indicates that large language models (LLMs) have demonstrated remark-able capabilities in various programming-related domains, such as code generation and code refinement.
Approach:	They propose a framework that combines exploration with refinement to reduce test-time computation overhead.
Outcome:	The proposed framework outperforms SOTA and AgentCoder on humanEval and MBPP benchmarks while reducing test-time computation overhead and scalability.

Evaluating Character Understanding of Large Language Models via Character Profiling from Fictional Works (2024.emnlp-main)

Copied to clipboard

Xinfeng Yuan, Siyu Yuan, Yuhan Cui, Tianhe Lin, Xintao Wang, Rui Xu, Jiangjie Chen, Deqing Yang

H-MAS: Hierarchical Multi-Agent Scheduling for Multi-Tenant LLM Serving (2026.findings-acl)

Challenge:	Recent advances in large language models (LLMs) have catalyzed numerous AI applications, among which role-playing agents (RPAs) are particularly popular.
Approach:	They propose to evaluate LLMs' character understanding capability via the character profiling task, i.e., summarizing character profiles from corresponding materials, a widely adopted yet understudied practice for RPA development.
Outcome:	The proposed model outperforms existing models and literature summarization methods and proves its ability to understand fictional characters in downstream tasks.

Copied to clipboard

Yuhan Liu, Cong Xu, Qi Jia, Yihua Wang, Feiyu Chen, Liang Jin, Lu Liu, Yaqian Zhao, Yuting Ding, Xiang Li

Challenge:	Multi-tenant Model-as-a-Service (MaaS) workloads exhibit non-stationarity across multiple time scales . existing request schedulers often rely on a fixed policy that remains unchanged at runtime .
Approach:	They propose a hierarchical multi-agent scheduler that operates in a layered closed loop . they propose to maintain 1.2–3.0 higher Goodput than SGLang and vLLM .
Outcome:	Experiments show that H-MAS achieves 1.2–3.0 higher Goodput than SGLang and vLLM . it maintains more stable QoS under diverse request lengths and heterogeneous SLO targets .

HoPE: A Novel Positional Encoding Without Long-Term Decay for Enhanced Context Awareness and Extrapolation (2025.acl-long)

Copied to clipboard

Yuhan Chen, Ang Lv, Jian Luan, Bin Wang, Wei Liu

MemRec: Collaborative Memory-Augmented Agentic Recommender System (2026.acl-long)

Challenge:	Existing positional encodings exhibit long-term decay, based on an entrenched and long-standing opinion that tokens farther away from the current position carry less relevant information.
Approach:	They propose a high-frequency rotary position encoding (HoPE) that replaces specific components in RoPE with position-independent ones, retaining only high- frequency signals.
Outcome:	The proposed method exhibits greater robustness to the out-of-distribution behavior in attention patterns during extrapolation.

Copied to clipboard

Weixin Chen, Yuhan Zhao, Jingyuan Huang, Zihe Ye, Mingxuan Ju, Tong Zhao, Neil Shah, Li Chen, Yongfeng Zhang

From Style to Story: A Curriculum Learning Approach for Imitative Novel Generation (2026.findings-acl)

Challenge:	Existing recommender systems rely on semantic user and item memories to make predictions, but these memories are kept in isolation.
Approach:	They propose a framework that architecturally decouples memory management from reasoning to decouple memory management and reasoning from the user and item memories.
Outcome:	The proposed framework decouples memory management from reasoning and achieves state-of-the-art performance on four benchmarks.

Copied to clipboard

Xueran Han, Yuhan Liu, Mingzhe Li, Wei Liu, Sen Hu, Rui Yan, Zhiqiang xu, Xiuying Chen

IAD: In-Context Learning Ability Decoupler of Large Language Models in Meta-Training (2024.lrec-main)

Challenge:	Novels create rich, immersive worlds with intricate plots and distinct styles, captivating readers through complex storytelling.
Approach:	They propose a novel generation system that imitates novel elements by predicting plot developments and writing concrete details using vivid, expressive language.
Outcome:	The novel imitative novel generation system is trained through a curriculum learning paradigm, progressing from low-level stylistic mastery to high-level narrative coherence.

Copied to clipboard

Yuhan Liu, Xiuying Chen, Gao Xing, Ji Zhang, Rui Yan

ShieldHead: Decoding-time Safeguard for Large Language Models (2025.findings-acl)

Challenge:	In-context Learning (ICL) is a paradigm in which LLMs acquire task-specific knowledge by processing input-output pairs provided as prompts.
Approach:	They propose an In-context learning Ability Decoupler to separate ICL ability from general ability of LLMs in meta-training phase . they first identify parameters suitable for ICL by transference-driven gradient importance and propose a new max-margin loss to emphasize the separation of the two abilities.
Outcome:	The proposed model separates the ICL ability from the general ability of LLMs in the meta-training phase, where the I-related parameters are tuned to adapt for ICL tasks.

Copied to clipboard

Zitao Xuan, Xiaofeng Mao, Da Chen, Xin Zhang, Yuhan Dong, Jun Zhou

Constructing coherent spatial memory in LLM agents through graph rectification (2026.acl-long)

Challenge:	Recent advances in LLM-based moderation methods have demonstrated remarkable promise in identifying safety risks associated with both inputs and outputs in human-AI interactions.
Approach:	They propose to learn a classification head on the last-layer hidden states of a dialogue model and use it to detect harmful content.
Outcome:	The proposed framework is 300 faster (1ms) than previous LLM-based moderation models with 99% less parameters than LlamaGuard.

Copied to clipboard

Puzhen Zhang, Xuyang Chen, Yu Feng, Yuhan Jiang, Liqiu Meng

Structured Data in NLP

Challenge:	Existing approaches to map construction and map repair are limited by context constraints and inconsistencies in text processing environments.
Approach:	They propose a framework for LLM-driven construction and map repair . it detects, localizes, and corrects structural inconsistencies in incremental navigation graphs .
Outcome:	The proposed method significantly improves map correctness and robustness in entangled or chained inconsistencies.

Looking Beyond the One: Operationalizing and Eliciting Visual Ambiguity in VLLMs (2026.acl-long)

Copied to clipboard

Yuchong Chen, Bowei Zou, Yuhan Chen, Yifan Fan, Xinyu Li, Shujun Cao, Yu Hong

Visual Data in NLP

Challenge:	Visual question answering systems typically collapse ambiguity, committing to a single interpretation during decoding and evaluation.
Approach:	They operationalize ambiguity as the existence of multiple answer-supporting regions in an image . they show that ambiguities are already encoded in their internal representations .
Outcome:	The proposed approach makes ambiguity observable without exhaustive annotations . ambiguities are already encoded in models, but not reliably expressed in outputs despite hidden states .

Rethinking Expert Trajectory Utilization in LLM Post-training for Mathematical Reasoning (2026.acl-long)

Copied to clipboard

Bowen Ding, Yuhan Chen, Jiayang Lyu, Jiyao Yuan, Qi Zhu, Shuangshuang Tian, Dantong Zhu, Futing Wang, Heyuan Deng, Fei Mi, Lifeng Shang, Tao Lin

Reasoning

Challenge:	evolving generic Large Language Models into specialized Large Reasoning Models requires effective post-training.
Approach:	They propose a plasticity-ceiling framework to harness expert trajectories . they establish the Sequential SFT-then-RL pipeline as the superior standard .
Outcome:	The proposed framework overcomes stability and premature convergence deficits in synchronized approaches.

More is not always better? Enhancing Many-Shot In-Context Learning with Differentiated and Reweighting Objectives (2025.acl-long)

Copied to clipboard

Xiaoqing Zhang, Ang Lv, Yuhan Liu, Flood Sung, Wei Liu, Jian Luan, Shuo Shang, Xiuying Chen, Rui Yan

Batch-ICL: Effective, Efficient, and Order-Agnostic In-Context Learning (2024.findings-acl)

Challenge:	Large language models excel at few-shot in-context learning but performance plateaus as ICL demonstrations increase from a few to many.
Approach:	They propose a novel optimization method that optimizes the negative log-likelihood objective and reweights the model to achieve many-shot performance.
Outcome:	The proposed method achieves significant performance improvements across a large-scale dataset.

Copied to clipboard

Kaiyi Zhang, Ang Lv, Yuhan Chen, Hansen Ha, Tao Xu, Rui Yan

GlossaGen: Making Academic Translation Smarter with Glossing (2026.findings-acl)

Challenge:	In-context learning (ICL) is a meta-optimization process that affects performance . we develop a batch-based inference algorithm that is order-agnostic to ICL examples .
Approach:	They develop an order-agnostic inference algorithm that aggregates ICL examples in batches . they find it outperforms most permutations of ICL, and it even exceeds the best order .
Outcome:	The proposed method outperforms standard ICL examples while reducing computational resources.

Copied to clipboard

Zixiao Wang, Duzhen Zhang, Juntian Zhang, Yuhan Liu, Guoming Li, Haolun Wu, Le Song, Xiuying Chen