Papers by Yuhui Zhang
EAGLE-2: Faster Inference of Language Models with Dynamic Draft Trees (2024.emnlp-main)
Copied to clipboard
| Challenge: | Modern Large Language Models (LLMs) are expensive and time-consuming. |
| Approach: | They propose a new technique of context-aware dynamic draft tree into drafting modeling. |
| Outcome: | The proposed method achieves speedup ratios of up to **5x**, which is 1.3x that of EAGLE. |
MAKAR: a Multi-Agent framework based Knowledge-Augmented Reasoning for Grounded Multimodal Named Entity Recognition (2025.emnlp-main)
Copied to clipboard
Xinkui Lin, Yuhui Zhang, Yongxiu Xu, Kun Huang, Hongzhang Mu, Yubin Wang, Gaopeng Gou, Li Qian, Li Peng, Wei Liu, Jian Luan, Hongbo Xu
| Challenge: | Existing methods for GMNER fail to address semantic ambiguity caused by polysemy and long-tail distribution of datasets. |
| Approach: | They propose a framework for Grounded Multimodal Named Entity Recognition that leverages a Multimodal Large Language Model to address semantic ambiguity. |
| Outcome: | Extensive experiments show that the proposed framework outperforms existing methods on two benchmark datasets. |
One QuantLLM for ALL: Fine-tuning Quantized LLMs Once for Efficient Deployments (2025.acl-long)
Copied to clipboard
| Challenge: | Quantization has shown promise for Large Language Models, but current methods require lengthy training to alleviate quantization loss. |
| Approach: | They propose to decouple weights and incorporate Low-Rank adapters to reduce weight sharing . they validate the approach on LLaMA2 families and Mistral on downstream evaluation . |
| Outcome: | The proposed approach shows high performance while reducing deployment time faced with multiple scenarios. |
PFDial: A Structured Dialogue Instruction Fine-tuning Method Based on UML Flowcharts (2025.findings-acl)
Copied to clipboard
Ming Zhang, Yuhui Wang, Yujiong Shen, Tingyi Yang, Changhao Jiang, Yilong Wu, Shihan Dou, Qinhao Chen, Zhiheng Xi, Zhihao Zhang, Yi Dong, Zhen Wang, Zhihui Fei, Mingyang Wan, Tao Liang, Guojun Ma, Qi Zhang, Tao Gui, Xuanjing Huang
| Challenge: | Large Language Models (LLMs) have shown remarkable progress in dialogue and reasoning, but they struggle to solve strictly constrained dialogue tasks. |
| Approach: | They construct a dataset that contains 12,705 high-quality Chinese dialogue instructions from 440 flowcharts containing 5,055 process nodes. |
| Outcome: | The proposed model outperforms GPT-4o models on backward transitions and outperformed GPT-42 models on the same dataset. |
NegVQA: Can Vision Language Models Understand Negation? (2025.findings-acl)
Copied to clipboard
| Challenge: | NegVQA is a visual question answering (VQA) benchmark consisting of 7,379 two-choice questions covering diverse negation scenarios and image-question distributions. |
| Approach: | They propose a visual question answering benchmark consisting of 7,379 two-choice questions covering diverse negation scenarios and image-question distributions. |
| Outcome: | The proposed model fails to correctly interpret negation, leading to critical errors in interactive AI systems. |
LLMEval-Fair: A Large-Scale Longitudinal Study on Robust and Fair Evaluation of Large Language Models (2026.acl-long)
Copied to clipboard
Ming Zhang, Yujiong Shen, Jingyi Deng, Yuhui Wang, Huayu Sha, Kexin Tan, Qiyuan Peng, Yue Zhang, Junzhe Wang, Shichun Liu, Yueyuan Huang, Jingqi Tong, Changhao Jiang, Yilong Wu, Zhihao Zhang, Mingqi Wu, Mingxu Chai, Zhiheng Xi, Shihan Dou, Tao Gui, Qi Zhang, Xuanjing Huang
| Challenge: | Existing evaluation of Large Language Models on static benchmarks is vulnerable to data contamination and leaderboard overfitting. |
| Approach: | LLMEval-Fair framework provides a framework for dynamic evaluation of Large Language Models . evaluators use a proprietary bank of 220k graduate-level questions to analyze model data . |
| Outcome: | LLMEval-Fair provides robust and credible evaluation framework for Large Language Models . it provides a strong empirical validation for the dynamic evaluation paradigm . |
Eureka: Neural Insight Learning for Knowledge Graph Reasoning (2022.coling-1)
Copied to clipboard
| Challenge: | Existing knowledge embedding methods have limited performance on knowledge graph reasoning tasks . eureka is empowered to learn seen relations with sufficient training triples . |
| Approach: | They propose a neural insight learning framework called Eureka to bridge the “seen” to “unsea” gap . Eureca is empowered to learn seen relations with sufficient training triples while providing flexibility to learn unseen relations given only one trigger . |
| Outcome: | The proposed framework outperforms state-of-the-art models on seen and unseen relations . it can learn seen and unseen relationships with sufficient training triples . |
Not All Experts are Equal: Efficient Expert Pruning and Skipping for Mixture-of-Experts Large Language Models (2024.acl-long)
Copied to clipboard
| Challenge: | Mixture-of-Experts (MoE) LLMs achieve higher performance with fewer active parameters, but are still difficult to deploy due to their immense parameter sizes. |
| Approach: | They propose expert-level sparsification techniques to enhance the deployment efficiency of large language models by introducing plug-and-play expert pruning and skipping techniques. |
| Outcome: | The proposed methods reduce model sizes and increase inference speed while maintaining satisfactory performance across a wide range of tasks. |
Beyond Positive Scaling: How Negation Impacts Scaling Trends of Language Models (2023.findings-acl)
Copied to clipboard
Yuhui Zhang, Michihiro Yasunaga, Zhengping Zhou, Jeff Z. HaoChen, James Zou, Percy Liang, Serena Yeung
| Challenge: | Recent studies show that some tasks exhibit inverse scaling, or U-shaped scaling, where the performance degrades as models are scaled up. |
| Approach: | They propose a task that asks questions with negation to show positive scaling . they hypothesize that solving NeQA depends on question answering and negation understanding . |
| Outcome: | The proposed task can exhibit inverse scaling, U-shaped scaling, or positive scaling, and the scaling trends shift as the task is more powerful. |
Chain-of-Talkers (CoTalk): Fast Human Annotation of Dense Image Captions (2025.emnlp-main)
Copied to clipboard
| Challenge: | Existing approaches for optimizing human annotation efforts are limited . et al., 2015) suggest that densely annotated image captions improve vision-language alignment . |
| Approach: | They propose an AI-in-the-loop methodology to maximize the number of annotated samples and improve their comprehensiveness under fixed budget constraints. |
| Outcome: | The proposed method improves annotation speed and retrieval performance over the parallel method. |
Inducing Grammar from Long Short-Term Memory Networks by Shapley Decomposition (2020.acl-srw)
Copied to clipboard
| Challenge: | a recent study shows that modern neural networks understand sentences implicitly by inducing recursive structures. |
| Approach: | They propose to explicitly induce grammar by tracing the computational process of a long short-term memory network. |
| Outcome: | The proposed model can explicitly induce grammar without external knowledge . tracing the computational process of a long short-term memory network is shown to be effective . |
Pre-trained Language Models Do Not Help Auto-regressive Text-to-Image Generation (2024.emnlp-main)
Copied to clipboard
| Challenge: | Recent advances in image tokenizers have enabled text-to-image generation using auto-regressive methods, but these methods lack pre-trained language models for text-based models. |
| Approach: | They adapt a pre-trained language model for auto-regressive text-to-image generation and show that pre-train language models offer limited help. |
| Outcome: | The proposed model is compared with a pre-trained language model and shows that it is no more effective than random initialized models. |
Do Current Video LLMs Have Strong OCR Abilities? A Preliminary Study (2025.coling-main)
Copied to clipboard
| Challenge: | a new benchmark evaluates video-based optical character recognition (Video OCR) performance of multi-modal models in videos . the benchmark aims to improve video LLMs' ability to extract text from video content . previous benchmarks have focused on video QA, but not video-related QA. |
| Approach: | They propose to evaluate the video OCR performance of multi-modal models in videos . they use a semi-automated approach that integrates the OCR ability of image LLMs with manual refinement . |
| Outcome: | The proposed benchmark includes 1,028 videos and 2,961 question-answer pairs . it integrates the OCR ability of image LLMs with manual refinement . |
LLMEval-Med: A Real-world Clinical Benchmark for Medical LLMs with Physician Validation (2025.findings-emnlp)
Copied to clipboard
Ming Zhang, Yujiong Shen, Zelin Li, Huayu Sha, Binze Hu, Yuhui Wang, Chenhao Huang, Shichun Liu, Jingqi Tong, Changhao Jiang, Mingxu Chai, Zhiheng Xi, Shihan Dou, Tao Gui, Qi Zhang, Xuanjing Huang
| Challenge: | Current medical benchmarks have limitations in question design, data sources and evaluation methods. |
| Approach: | They propose a new benchmark covering five core medical areas . it includes 2,996 questions created from real-world electronic health records . |
| Outcome: | The proposed model covers five core medical areas and includes 2,996 questions created from real-world electronic health records and expert-designed clinical scenarios. |
Parrot: A Training Pipeline Enhances Both Program CoT and Natural Language CoT for Reasoning (2025.emnlp-main)
Copied to clipboard
Senjie Jin, Lu Chen, Zhiheng Xi, Yuhui Wang, Sirui Song, Yuhao Zhou, Xinbo Zhang, Peng Sun, Hong Lu, Tao Gui, Qi Zhang, Xuanjing Huang
| Challenge: | Existing work focuses on enabling models to generate natural language chain-of-thought rationales or leverage executable and verifiable code, such as Python. |
| Approach: | They propose a novel training pipeline that integrates sequential P-CoT and N-Co T generation and a subtask hybrid training strategy to facilitate natural language transferability. |
| Outcome: | The proposed training pipeline improves both N-CoT and P-Co T performance over the RL baseline. |
Jiuge: A Human-Machine Collaborative Chinese Classical Poetry Generation System (P19-3)
Copied to clipboard
Guo Zhipeng, Xiaoyuan Yi, Maosong Sun, Wenhao Li, Cheng Yang, Jiannan Liang, Huimin Chen, Yuhui Zhang, Ruoyu Li
| Challenge: | Existing systems for automatic poetry generation are model-oriented, resulting in poor user participation. |
| Approach: | They propose a human-machine collaborative Chinese classical poetry generation system called Jiuge . Jiuge allows users to revise unsatisfied parts of a generated poem draft repeatedly . |
| Outcome: | The proposed system allows users to revise unsatisfied parts of a generated poem draft repeatedly. |
PaperSearchQA: Learning to Search and Reason over Scientific Papers with RLVR (2026.eacl-long)
Copied to clipboard
James Burgess, Jan N. Hansen, Duo Peng, Yuhui Zhang, Alejandro Lozano, Min Woo Sun, Emma Lundberg, Serena Yeung-Levy
| Challenge: | Recent methods supervise only the final answer accuracy using reinforcement learning with verifiable rewards (RLVR). |
| Approach: | They propose to train search agents to search and reason over scientific papers and a factoid QA dataset with 60k biomedical paper abstracts. |
| Outcome: | The proposed model outperforms non-RL retrieval baselines and is scalable and extendable to other scientific domains. |
VRPO: Rethinking Value Modeling for Robust RL under Noisy Supervision in LLM Post-Training (2026.acl-long)
Copied to clipboard
Dingwei Zhu, Shihan Dou, Zhiheng Xi, Senjie Jin, Guoqiang Zhang, Jiazheng Zhang, Junjie Ye, Mingxu Chai, Enyu Zhou, Ming Zhang, Yuhui Wang, Caishuang Huang, Chenhao Huang, Yunke Zhang, Yuran Wang, Tao Gui, Qi Zhang, Xipeng Qiu, Xuanjing Huang
| Challenge: | Reinforcement Learning (RL) in real-world environments often suffers from ambiguous or incomplete supervision. |
| Approach: | They propose a framework that enhances value modeling for robust RL in LLM post-training by integrating auxiliary losses guided by entropy and perplexity from a frozen language model and variational information bottleneck. |
| Outcome: | The proposed framework outperforms baselines on multi-turn dialogue, math reasoning, and science QA with rule-based and model-based rewards. |
OAgents: An Empirical Study of Building Effective Agents (2025.findings-emnlp)
Copied to clipboard
He Zhu, Tianrui Qin, King Zhu, Heyuan Huang, Yeyi Guan, Jinxiang Xia, Hanhao Li, Yi Yao, Ningning Wang, Pai Liu, Tianhao Peng, Xin Gui, Li Xiaowan, Yuhui Liu, Xiangru Tang, Jian Yang, Ge Zhang, Xitong Gao, Yuchen Eleanor Jiang, Changwang Zhang, Jun Wang, Jiaheng Liu, Wangchunshu Zhou
| Challenge: | a recent study shows that agent research practices are far from standard, rigorous . lack of a standard evaluation protocol makes previous works not reproducible, authors say . |
| Approach: | They conduct an empirical study on the GAIA benchmark to investigate agent design choices . they find that lack of a standard evaluation protocol makes previous works not reproducible . |
| Outcome: | The proposed framework achieves state-of-the-art performance among open-source projects. |
Data or Language Supervision: What Makes CLIP Better than DINO? (2025.findings-emnlp)
Copied to clipboard
| Challenge: | CLIP outperforms self-supervised models like DINO as vision encoders for vision-language models (VLMs) but it remains unclear whether this advantage stems from CLIP’s language supervision or its much larger training data. |
| Approach: | Embedding analysis shows CLIP captures high-level semantics while DINO is more responsive to low-level features like colors and styles. |
| Outcome: | Embedding analysis shows that CLIP captures high-level semantics, while DINO is more responsive to low-level features like colors and styles. |
EquiBench: Benchmarking Large Language Models’ Reasoning about Program Semantics via Equivalence Checking (2025.emnlp-main)
Copied to clipboard
Anjiang Wei, Jiannan Cao, Ran Li, Hongyu Chen, Yuhui Zhang, Ziheng Wang, Yuan Liu, Thiago S. F. X. Teixeira, Diyi Yang, Ke Wang, Alex Aiken
| Challenge: | EquiBench is a new benchmark to evaluate large language models' ability to reason about program semantics . Unlike natural language, code is executable. |
| Approach: | They propose a benchmark to evaluate large language models through equivalence checking . EquiBench consists of 2400 program pairs across four languages and six categories . |
| Outcome: | The proposed benchmark consists of 2400 program pairs across four languages and six categories. |
Stanza: A Python Natural Language Processing Toolkit for Many Human Languages (2020.acl-demos)
Copied to clipboard
| Challenge: | Existing tools that support only a few major languages are under-optimized for accuracy due to a focus on efficiency or use of less powerful models. |
| Approach: | They introduce a Python natural language processing toolkit that supports 66 languages . they train Stanza on 112 datasets and show it generalizes well on all languages compared to other tools . |
| Outcome: | The proposed toolkit performs well on 112 datasets and is compatible with the popular Java CoreNLP software. |
ERNIE-Layout: Layout Knowledge Enhanced Pre-training for Visually-rich Document Understanding (2022.findings-emnlp)
Copied to clipboard
Qiming Peng, Yinxu Pan, Wenjin Wang, Bin Luo, Zhenyu Zhang, Zhengjie Huang, Yuhui Cao, Weichong Yin, Yongfeng Chen, Yin Zhang, Shikun Feng, Yu Sun, Hao Tian, Hua Wu, Haifeng Wang
| Challenge: | Existing methods for visually rich document understanding lack layout-centered knowledge . experimental results show that ERNIE-Layout improves layout awareness . |
| Approach: | They propose a document pre-training solution with layout knowledge enhancement in the whole workflow to learn better representations that combine the features from text, layout, and image. |
| Outcome: | The proposed model outperforms existing models on key downstream tasks. |