Papers by Chenyang Wang
ReFreeKV: Towards Threshold-Free KV Cache Compression (2026.findings-acl)
Copied to clipboard
| Challenge: | Towards the KV cache efficiency, we propose a new objective that lifts the threshold constraints for robust KV compression. |
| Approach: | They propose a method that adjusts KV cache budgets while preserving full-cache performance. |
| Outcome: | The proposed method can reduce memory consumption while preserving full-cache performance. |
Disentangle to Decay: Linear Attention with Trainable Decay Factor (2025.coling-main)
Copied to clipboard
| Challenge: | Existing linear attention models use a decay factor based positional encoding (PE), but the decay factor is manually designed and non-trainable, limiting further optimization. |
| Approach: | They propose a PE-based positional encoding that disentangles decay factor into two parts to achieve further optimization and stable training. |
| Outcome: | The proposed model achieves stable training of decay factor and improves inference efficiency in normal context and extrapolation scenarios. |
Out-of-Distribution Generalization in Natural Language Processing: Past, Present, and Future (2023.emnlp-main)
Copied to clipboard
Linyi Yang, Yaoxian Song, Xuan Ren, Chenyang Lyu, Yidong Wang, Jingming Zhuo, Lingqiao Liu, Jindong Wang, Jennifer Foster, Yue Zhang
| Challenge: | Existing literature on the generalization of machine learning models to out-of-distribution data is lacking. |
| Approach: | They propose to present the first comprehensive review of recent progress, methods, and evaluations on the generalization challenge from an OOD perspective in natural language understanding. |
| Outcome: | The proposed survey provides the first comprehensive review of recent progress, methods, and evaluations on the generalization challenge from an OOD perspective in natural language understanding. |
AgentGym: Evaluating and Training Large Language Model-based Agents across Diverse Environments (2025.acl-long)
Copied to clipboard
Zhiheng Xi, Yiwen Ding, Wenxiang Chen, Boyang Hong, Honglin Guo, Junzhe Wang, Xin Guo, Dingwen Yang, Chenyang Liao, Wei He, Songyang Gao, Lu Chen, Rui Zheng, Yicheng Zou, Tao Gui, Qi Zhang, Xipeng Qiu, Xuanjing Huang, Zuxuan Wu, Yu-Gang Jiang
| Challenge: | Large language models (LLMs) are promising foundations to build generally-capable agents . however, the community lacks a unified interactive framework that covers diverse environments for comprehensive evaluation of agents. |
| Approach: | They propose a framework that features 7 real-world scenarios, 14 environments, and 89 tasks for unified, real-time, and concurrent agent interaction. |
| Outcome: | The proposed framework features 7 real-world scenarios, 14 environments, and 89 tasks for unified, real-time, and concurrent agent interaction. |
Learn from Failure: Fine-tuning LLMs with Trial-and-Error Data for Intuitionistic Propositional Logic Proving (2024.acl-long)
Copied to clipboard
Chenyang An, Zhibo Chen, Qihao Ye, Emily First, Letian Peng, Jiayun Zhang, Zihan Wang, Sorin Lerner, Jingbo Shang
| Challenge: | Recent advances in Automated Theorem Proving have shown the effectiveness of leveraging a (large) language model that generates tactics (i.e. proof steps) to search through proof states. |
| Approach: | They propose to use a large language model that generates tactics to search through proof states. |
| Outcome: | The proposed model solves more unseen theorems with lower trial searches than the current model, which only learns from failed attempts. |
Marco-Bench-MIF: On Multilingual Instruction-Following Capability of Large Language (2025.acl-long)
Copied to clipboard
Bo Zeng, Chenyang Lyu, Sinuo Liu, Mingyan Zeng, Minghao Wu, Xuanfan Ni, Tianqi Shi, Yu Zhao, Yefeng Liu, Chenyu Zhu, Ruizhe Li, Jiahui Geng, Qing Li, Yu Tong, Longyue Wang, Weihua Luo, Kaifu Zhang
| Challenge: | Existing datasets for instruction-following are monolingual and centered on English . existing data are unable to capture linguistic and cultural subtle differences . |
| Approach: | They propose an extension of IFEval to a localized multilingual version called Marco-Bench-MIF . their benchmark addresses linguistic constraints and cultural references via translation and verification . |
| Outcome: | The proposed extension of IFEval to a localized multilingual version covers 30 languages with varying levels of localization. |
MM-ShiftKV: Decode-Aware Prefill-Stage KV Selection for Multimodal Large Language Models (2026.findings-acl)
Copied to clipboard
| Challenge: | Recent work suggests a prefill-stage KV cache selection method to estimate KV importance from prefilling statistics. |
| Approach: | They propose a training-free, decode-aware and strictly prefill-only KV selection method that retains key-value caching for decoding . |
| Outcome: | The proposed method outperforms existing methods under tight cache budgets on multimodal benchmarks. |
CHROMIC: Chronological Reasoning Across Multi-Panel Comics (2026.eacl-long)
Copied to clipboard
Bingxuan Hou, Jiayi Lin, Chenyang Zhang, Dapeng Yin, Shuyue Zhu, Qingqing Hong, Mengna Gao, Junli Wang
| Challenge: | Large-scale vision–language models have achieved remarkable progress on various reasoning tasks, but most studies focus on natural photographic images and pay limited attention to multi-panel visual narratives such as comics. |
| Approach: | They propose a benchmark dataset for chronological reasoning in multi-panel comics that covers six types of reasoning questions and spans both Western and Japanese comic styles. |
| Outcome: | The proposed dataset covers six types of reasoning questions and spans both Western and Japanese comic styles. |
Benchmarking and Improving Long-Text Translation with Large Language Models (2024.findings-acl)
Copied to clipboard
Longyue Wang, Zefeng Du, Wenxiang Jiao, Chenyang Lyu, Jianhui Pang, Leyang Cui, Kaiqiang Song, Derek Wong, Shuming Shi, Zhaopeng Tu
| Challenge: | Recent studies have illuminated the promising capabilities of large language models (LLMs) in handling long texts. |
| Approach: | They construct a benchmark dataset specifically designed for the finetuning and evaluation of large language models (LLMs) they compare LLMs with MT models and find they exhibit shortcomings in long-text domains . |
| Outcome: | The proposed model performs better in long-text translation, and its performance diminishes as document size increases. |
Document-Level Machine Translation with Large Language Models (2023.emnlp-main)
Copied to clipboard
| Challenge: | Large language models (LLMs) such as ChatGPT can produce coherent, cohesive, relevant, and fluent answers for various natural language processing tasks. |
| Approach: | They examine the impact of different prompts on document-level translation quality and discourse phenomena using figures and lines, which are invisible to GPT-4. |
| Outcome: | The proposed models outperform commercial MT systems and advanced document-level MT methods on a number of benchmarks and show potential to become a new paradigm for document- level translation. |
RT-VC: Real-Time Zero-Shot Voice Conversion with Speech Articulatory Coding (2025.acl-demo)
Copied to clipboard
| Challenge: | Experimental evaluations show RT-VC delivers a 13.3% reduction in latency . voice conversion modifies speech to match the timbre of a target speaker while preserving content information. |
| Approach: | They propose a zero-shot real-time voice conversion system that leverages an articulatory feature space to naturally disentangle content and speaker characteristics. |
| Outcome: | The proposed system achieves a CPU latency of 61.4 ms, representing a 13.3% reduction in latency. |
Adversarial Preference Learning for Robust LLM Alignment (2025.findings-acl)
Copied to clipboard
Yuanfu Wang, Pengyu Wang, Chenyang Xi, Bo Tang, Junyi Zhu, Wenqiang Wei, Chen Chen, Chao Yang, Jingfeng Zhang, Chaochao Lu, Yijun Niu, Keming Mao, Zhiyu Li, Feiyu Xiong, Jie Hu, Mingchuan Yang
| Challenge: | Modern language models rely on Reinforcement Learning from Human Feedback (RLHF) to encourage safe behaviors, but they remain vulnerable to adversarial attacks due to three key limitations: (1) the inefficiency and high cost of human annotation; (2) the vast diversity of potential adversarials; and (3) the risk of feedback bias and reward hacking. |
| Approach: | They propose an iterative adversarial training method that incorporates three key innovations to address these challenges. |
| Outcome: | Experiments on Mistral-7B-Instruct-v0.3 show that the proposed method significantly enhances robustness and reduces harmful outputs from 5.88% to 0.43%. |
The Price of Format: Diversity Collapse in LLMs (2025.findings-emnlp)
Copied to clipboard
| Challenge: | Instruction-tuned large language models employ structured templates to enforce format consistency during inference. |
| Approach: | They fine-tune instruction-tuning large language models with structured templates and evaluate their results across three axes: downstream task performance, alignment behavior, and output diversity. |
| Outcome: | The proposed model generates semantically similar outputs even under high temperature sampling and structural tokens in templates significantly constrain the model’s output space. |
Correct after Answer: Enhancing Multi-Span Question Answering with Post-Processing Method (2024.findings-emnlp)
Copied to clipboard
| Challenge: | Prior work focuses on designing specific methods or applying heuristic strategies to encourage models to predict more correct predictions. |
| Approach: | They propose a framework that uses a post-processing strategy to handle incorrect predictions. |
| Outcome: | The proposed framework significantly improves the Exact Match scores on multiple MSQA datasets. |
Detecting AI-Generated Content on Social Media with Multi-modal Language Models (2026.acl-industry)
Copied to clipboard
Chenyang Yang, Shen Yan, Yibo Yang, Litao Hu, Yuchen Liu, Yuan Zeng, Hanchao Yu, Yinan Zhu, Sumedha Singla, Brian Vanover, Huijun Qian, Zihao Wang, Fujun Liu, Aashu Singh, Jianyu Wang, Xuewen Zhang
| Challenge: | Existing methods for AI-generated content detection face poor generalization to newer models, reliance on single modalities, and lack of interpretable explanations. |
| Approach: | They propose a model that curates diverse social media data and trains a vision-language model for detection and explanation. |
| Outcome: | The proposed model achieves state-of-the-art detection performance on public benchmarks and observes positive downstream impacts on user engagement. |
CURA: Clinical Uncertainty Risk Alignment for Language Model–Based Risk Prediction (2026.acl-long)
Copied to clipboard
| Challenge: | Clinical language models (LMs) are increasingly applied to support clinical risk prediction from free-text notes, yet their uncertainty estimates are poorly calibrated and clinically unreliable. |
| Approach: | They propose a framework that aligns clinical LM-based risk estimates and uncertainty with individual error likelihoods and cohort-level ambiguities. |
| Outcome: | The proposed framework improves accuracy on clinical risk prediction tasks without compromising discrimination. |
On the Representation Geometry of LoRA Model Merging (2026.findings-acl)
Copied to clipboard
| Challenge: | Existing methods for low-rank Adaptation (LoRA) fine-tuning focus on globally shared structure . combining SVD with CUR improves performance of LoRA model merging . |
| Approach: | They propose a training-free method that combines SVD and CUR decomposition to improve LoRA merging performance. |
| Outcome: | The proposed procedure improves on vision and language benchmarks. |
Targeted Exploration via Unified Entropy Control for Reinforcement Learning (2026.findings-acl)
Copied to clipboard
| Challenge: | Existing methods for group relative policy optimization suffer from entropy collapse . Existing exploration methods introduce additional bias or variance during exploration, making it difficult to maintain stability. |
| Approach: | They propose a framework that provides targeted mechanisms for exploration and stabilization. |
| Outcome: | The proposed framework expands search space on difficult prompts while preventing entropy growth uncontrollably. |
VSCBench: Bridging the Gap in Vision-Language Model Safety Calibration (2025.findings-acl)
Copied to clipboard
Jiahui Geng, Qing Li, Zongxiong Chen, Yuxia Wang, Derui Zhu, Zhuohan Xie, Chenyang Lyu, Xiuying Chen, Preslav Nakov, Fakhri Karray
| Challenge: | Existing safety calibration methods focus on model undersafety, where the model responds to hazardous queries, while neglecting oversafetiness, where models refuse to answer safe queries. |
| Approach: | They propose safety calibration which addresses both undersafety and oversafetiness by comparing model responses to a novel dataset of 3,600 image-text pairs. |
| Outcome: | The proposed methods have been used to evaluate safety calibration across image-centric and text-centric scenarios. |
HD-NDEs: Neural Differential Equations for Hallucination Detection in LLMs (2025.acl-long)
Copied to clipboard
| Challenge: | Hallucination is a significant challenge for large language models, but current methods struggle when non-factual information arises in the early or mid-sequence of outputs, reducing their reliability. |
| Approach: | They propose a method that captures the full dynamics of large language models by using neural differential equations to assess the truthfulness of statements. |
| Outcome: | The proposed method achieves 14% improvement in AUC-ROC on the True-False dataset compared to state-of-the-art methods. |
PAR2-RAG: Planned Active Retrieval and Reasoning for Multi-Hop Question Answering (2026.acl-industry)
Copied to clipboard
| Challenge: | Multi-hop question answering is a practical bottleneck in industry applications . large language models (LLMs) fail frequently when evidence coverage is incomplete or reasoning trajectories drift . |
| Approach: | They propose a training-free two-stage framework that separates coverage from commitment . it performs breadth-first anchoring to build a high-recall evidence frontier . compared with IRCoT, it achieves 23.5% higher answer accuracy . |
| Outcome: | The proposed framework outperforms strong baselines in MHQA benchmarks and achieves 23.5% higher answer accuracy and 10.5% NDCG gains in retrieval quality. |
SeedBench: A Multi-task Benchmark for Evaluating Large Language Models in Seed Science (2025.acl-long)
Copied to clipboard
Jie Ying, Zihong Chen, Zhefan Wang, Wanli Jiang, Chenyang Wang, Zhonghang Yuan, Haoyang Su, Huanjun Kong, Fan Yang, Nanqing Dong
| Challenge: | Seed science is essential for modern agriculture, but its application in seed science remains limited due to a shortage of experts and limited availability of online resources. |
| Approach: | They evaluate 26 leading large language models and compare them against a set of benchmarks . they find that there is a gap between the power of LLMs and real-world seed science problems . |
| Outcome: | The new seed benchmark highlights the gap between the power of large language models and real-world seed science problems. |
Multi-Grained Knowledge Distillation for Named Entity Recognition (2021.naacl-main)
Copied to clipboard
| Challenge: | Pre-trained big models have delivered top performance in Seq2seq modeling, but their deployments in real-world applications are often hindered by excessive computations and memory demands. |
| Approach: | They propose a distillation scheme to efficiently transfer knowledge from big models to their cheaper counterparts. |
| Outcome: | The proposed scheme maximizes the assimilation of knowledge from the teacher model to the student model. |
Improving Text Generation with Student-Forcing Optimal Transport (2020.emnlp-main)
Copied to clipboard
Jianqiao Li, Chunyuan Li, Guoyin Wang, Hao Fu, Yuhchen Lin, Liqun Chen, Yizhe Zhang, Chenyang Tao, Ruiyi Zhang, Wenlin Wang, Dinghan Shen, Qian Yang, Lawrence Carin
| Challenge: | Maximum likelihood estimation (MLE) is used to train models, but during testing, the model is conditioned on previously generated tokens, resulting in exposure bias. |
| Approach: | They propose to use optimal transport to match the sequences generated in MLE and test modes to reduce exposure bias. |
| Outcome: | The proposed method is validated on machine translation, text summarization, and text generation tasks. |
Marco-o1 v2: Towards Widening The Distillation Bottleneck for Reasoning Models (2025.acl-long)
Copied to clipboard
Huifeng Yin, Yu Zhao, Minghao Wu, Xuanfan Ni, Bo Zeng, Huaiyu.wh Huaiyu.wh, Tianqi Shi, Liangying Shao, Chenyang Lyu, Longyue Wang, Weihua Luo, Kaifu Zhang
| Challenge: | Recent efforts to distill large reasoning models into smaller lightweight models have shown competitive performances. |
| Approach: | They propose to distill long Chain-of-Thought data to improve SFT and RL methods by constructing data from scratch using Monte Carlo Tree Search. |
| Outcome: | The proposed method significantly improves reasoning performance on various benchmarks such as math (GSM8K, MATH, AIME). |
Bold Claims or Self-Doubt? Factuality Hallucination Type Detection via Belief State (2025.findings-emnlp)
Copied to clipboard
| Challenge: | Existing studies focus on detecting the presence of hallucinations but lack a systematic classification approach, which hinders deeper exploration of their characteristics. |
| Approach: | They propose a method to categorize hallucinations into two types: Overconfident and Unaware . |
| Outcome: | The proposed method categorizes factuality hallucination into two types: Overconfident and Unaware Hallucinations. |
Beyond Sentence-level Labels: Integrating Conversational Context and Personal Experience for Natural Emotional Expression (2026.findings-acl)
Copied to clipboard
Haiyang Sun, Chenyang Le, Wei Wang, Leying Zhang, Chuang Li, Bing Han, Chenda Li, Mengxiao Bi, Yanmin Qian
| Challenge: | Existing systems rely on sentence-level labels, which fails to capture the subtle nuances of human affect. |
| Approach: | They propose to use a large-scale, context-aware speech corpus derived from multi-speaker audiobooks to generate a speech that is human-like. |
| Outcome: | The proposed model outperforms existing methods in terms of emotional expression accuracy and naturalness. |
Improving Textual Network Embedding with Global Attention via Optimal Transport (P19-1)
Copied to clipboard
Liqun Chen, Guoyin Wang, Chenyang Tao, Dinghan Shen, Pengyu Cheng, Xinyuan Zhang, Wenlin Wang, Yizhe Zhang, Lawrence Carin
| Challenge: | Existing methods for learning textual network embeddings are noisy and sparse. |
| Approach: | They propose to use text-based attention parsing to learn context-aware network embeddings. |
| Outcome: | The proposed model outperforms state-of-the-art methods in a number of domains. |
A Paradigm Shift: The Future of Machine Translation Lies with Large Language Models (2024.lrec-main)
Copied to clipboard
Chenyang Lyu, Zefeng Du, Jitao Xu, Yitao Duan, Minghao Wu, Teresa Lynn, Alham Fikri Aji, Derek F. Wong, Longyue Wang
| Challenge: | Large Language Models (LLMs) are introducing a new phase in machine translation . despite advances in MT, there are still many challenges to overcome . |
| Approach: | They propose to highlight several new directions for MT that are influenced by Large Language Models like GPT-4 and ChatGPT. |
| Outcome: | The proposed models offer vast linguistic understandings and bring innovative methodologies, such as prompt-based techniques, that have the potential to further elevate MT. |
cAST: Enhancing Code Retrieval-Augmented Generation with Structural Chunking via Abstract Syntax Tree (2025.findings-emnlp)
Copied to clipboard
| Challenge: | Existing line-based chunking heuristics often break semantic structures, splitting functions or merging unrelated code. |
| Approach: | They propose a structure-aware method that breaks large AST nodes into smaller chunks . this method generates self-contained, semantically coherent units across programming languages . |
| Outcome: | The proposed method boosts Recall@5 by 4.3 points on RepoEval retrieval and Pass@1 by 2.67 points on SWE-bench generation. |