Papers by Yiming Zhou
On the Sentence Embeddings from Pre-trained Language Models (2020.emnlp-main)
Copied to clipboard
| Challenge: | Pre-trained contextual representations like BERT have been widely used for NLP tasks. |
| Approach: | They propose to transform anisotropic sentence embedding distribution to smooth and isotropic Gaussian distribution by normalizing flows that are learned with an unsupervised objective. |
| Outcome: | The proposed method achieves significant performance gains over state-of-the-art embeddings on a variety of semantic textual similarity tasks. |
Extracting Topics with Simultaneous Word Co-occurrence and Semantic Correlation Graphs: Neural Topic Modeling for Short Texts (2021.findings-emnlp)
Copied to clipboard
| Challenge: | Empirical results validate that DWGTM can generate more semantically coherent topics than baseline topic models. |
| Approach: | They develop a neural topic model which extracts topics from word co-occurrence graphs . Empirical results validate that DWGTM can generate more semantically coherent topics than baseline topic models. |
| Outcome: | Empirical results show that the proposed model can generate more coherent topics than baseline topic models. |
DARM: Distribution-Aware Reward Modeling by Alleviating Biases from Low Preference-Context Dependency Data (2026.acl-long)
Copied to clipboard
Shaofan Liu, Guoqiang Zhang, Shihan Dou, Huiyuan Zheng, Yiming Zhou, Junjie Ye, Shaowen Wang, Shichun Liu, Jiazheng Zhang, Tao Gui, Qi Zhang, Xuanjing Huang
| Challenge: | Existing methods for training reward models are vulnerable to context neglect and degraded accuracy. |
| Approach: | They propose distribution-aware reward modeling that augments the RM objective with a conditional mutual information regularizer that maximizes context and the predicted reward conditioned on the response. |
| Outcome: | The proposed model improves performance in RLHF and improves accuracy in other settings. |
Rethinking RL Evaluation: Can Benchmarks Truly Reveal Failures of RL Methods? (2026.findings-acl)
Copied to clipboard
| Challenge: | Existing benchmarks for reinforcement learning for large language models do not accurately assess generalization. |
| Approach: | They propose three core principles for designing more faithful benchmarks: sufficient difficulty, balanced evaluation, and distributional robustness. |
| Outcome: | The proposed benchmarks do not accurately assess generalization across distribution shifts, difficulty levels, and counterfactual scenarios. |
LongCLI-Bench: A Preliminary Benchmark and Study for Long-horizon Agentic Programming in Command-Line Interfaces (2026.findings-acl)
Copied to clipboard
Yukang Feng, Jianwen Sun, Zelai Yang, Jiaxin Ai, Chuanhao Li, Zizhen Li, Fanrui Zhang, Kang He, Rui Ma, Jifan Lin, Jie Sun, Yang Xiao, Sizhuo Zhou, Wenxiao Wu, Yiming Liu, Pengfei Liu, Shenglin Zhang, Kaipeng Zhang
| Challenge: | Existing benchmarks for agentic programming in long-horizon command-line interface tasks are limited by short task horizons, data contamination from GitHub scraping, and a lack of fine-grained evaluation metrics. |
| Approach: | They propose a benchmark to evaluate agentic capabilities across long-horizon command-line interface tasks. |
| Outcome: | The proposed benchmarks cover four engineering categories: from scratch, feature addition, bug fixing, and refactoring. |
FLamE: Few-shot Learning from Natural Language Explanations (2023.acl-long)
Copied to clipboard
| Challenge: | Recent work has shown limited utility of natural language explanations in improving classification. |
| Approach: | They propose a two-stage few-shot learning framework that generates explanations and fine-tunes a smaller model with generated explanations. |
| Outcome: | The proposed framework increases inference accuracy over strong baselines, but human evaluation reveals that the majority of generated explanations does not adequately justify classification decisions. |
PACE: Prefix-Protected and Difficulty-Aware Compression for Efficient Reasoning (2026.findings-acl)
Copied to clipboard
Ruixiang Feng, Yuntao Wen, Silin Zhou, Ke Shi, Yifan Wang, Ran Le, Zhenwei An, Zongchao Chen, Chen Yang, Guangyue Peng, Yiming Jia, Dongsheng Wang, Tao Zhang, Lisi Chen, Yang Song, Shen Gao, Shuo Shang
| Challenge: | Existing LRMs often suffer from "overthinking" and excessively long reasoning traces . a dual-level framework for length compression of LRM is proposed . |
| Approach: | They propose a framework for prefix-protected and difficulty-aware compression under hierarchical supervision. |
| Outcome: | The proposed framework reduces token usage while improving accuracy on math benchmarks. |
CLUE: A Chinese Language Understanding Evaluation Benchmark (2020.coling-main)
Copied to clipboard
Liang Xu, Hai Hu, Xuanwei Zhang, Lu Li, Chenjie Cao, Yudong Li, Yechen Xu, Kai Sun, Dian Yu, Cong Yu, Yin Tian, Qianqian Dong, Weitang Liu, Bo Shi, Yiming Cui, Junyi Li, Jun Zeng, Rongzhao Wang, Weijian Xie, Yanting Li, Yina Patterson, Zuoyu Tian, Yiwen Zhang, He Zhou, Shaoweihua Liu, Zhe Zhao, Qipeng Zhao, Cong Yue, Xinrui Zhang, Zhengliang Yang, Kyle Richardson, Zhenzhong Lan
| Challenge: | Existing language evaluation benchmarks for English are limited to English . lack of such benchmarks makes it difficult to replicate success in other languages . |
| Approach: | They introduce a large-scale Chinese language understanding evaluation benchmark . the benchmark uses a set of current state-of-the-art pre-trained Chinese models . |
| Outcome: | The first large-scale Chinese Language Understanding Evaluation (CLUE) benchmark is released . the benchmark evaluates models across a wide range of tasks on original Chinese text . existing language evaluation benchmarks are mostly limited to English . |
CogBench: Benchmarking Cognitive Alignment of Large Language Models in Educational Question Answering (2026.findings-acl)
Copied to clipboard
| Challenge: | Large language models (LLMs) possess strong capabilities in language understanding and generation, as well as remarkable problem-solving abilities. |
| Approach: | They propose a benchmark to assess the cognitive alignment capabilities of large language models in educational QA. |
| Outcome: | The proposed evaluation benchmark assesses the cognitive alignment capabilities of large language models in educational QA. |
RAM-SD: Retrieval-Augmented Multi-agent framework for Sarcasm Detection (2026.acl-long)
Copied to clipboard
| Challenge: | Existing approaches to sarcastic detection use a uniform reasoning strategy . existing approaches lack a framework to deal with the diverse analytical demands of sarcasm . |
| Approach: | They propose a Retrieval-Augmented Multi-Agent framework for Sarcasm Detection . the framework provides transparent and interpretable reasoning traces . |
| Outcome: | The proposed framework outperforms existing methods on four benchmarks and outperformed the strong GPT-4o+CoC baseline. |
CIF-Bench: A Chinese Instruction-Following Benchmark for Evaluating the Generalizability of Large Language Models (2024.findings-acl)
Copied to clipboard
Yizhi Li, Ge Zhang, Xingwei Qu, Jiali Li, Zhaoqun Li, Noah Wang, Hao Li, Ruibin Yuan, Yinghao Ma, Kai Zhang, Wangchunshu Zhou, Yiming Liang, Lei Zhang, Lei Ma, Jiajun Zhang, Zuowen Li, Wenhao Huang, Chenghua Lin, Jie Fu
| Challenge: | a recent study shows that large language models have limited generalization in low-resource languages like Chinese. |
| Approach: | They propose to evaluate the zero-shot generalizability of large language models to the Chinese language . they release only half of the dataset publicly, with the remainder kept private . |
| Outcome: | The Chinese Instruction-Following Benchmark evaluates the generalizability of LLMs to the Chinese language. |
MobileBERT: a Compact Task-Agnostic BERT for Resource-Limited Devices (2020.acl-main)
Copied to clipboard
| Challenge: | Empirical studies show that MobileBERT is 4.3x smaller and 5.5x faster than BERT_BASE . BERT is one of the largest models ever in NLP, but suffers from heavy model size and high latency . |
| Approach: | They propose a tool to compress and accelerate the popular BERT model by task-agnostic application. |
| Outcome: | The proposed model is 4.3x smaller and 5.5x faster than BERT_BASE . it achieves competitive results on well-known benchmarks . |
ChatMusician: Understanding and Generating Music Intrinsically with LLM (2024.findings-acl)
Copied to clipboard
Ruibin Yuan, Hanfeng Lin, Yi Wang, Zeyue Tian, Shangda Wu, Tianhao Shen, Ge Zhang, Yuhang Wu, Cong Liu, Ziya Zhou, Liumeng Xue, Ziyang Ma, Qin Liu, Tianyu Zheng, Yizhi Li, Yinghao Ma, Yiming Liang, Xiaowei Chi, Ruibo Liu, Zili Wang, Chenghua Lin, Qifeng Liu, Tao Jiang, Wenhao Huang, Wenhu Chen, Jie Fu, Emmanouil Benetos, Gus Xia, Roger Dannenberg, Wei Xue, Shiyin Kang, Yike Guo
| Challenge: | Despite LLMs' impressive capabilities in musical knowledge, music reasoning remains an unsolved task. |
| Approach: | They propose an open-source large language model (LLM) that integrates intrinsic musical abilities into LLaMA2 and GPT-3.5. |
| Outcome: | The proposed model can understand and generate music with a pure text tokenizer without external multi-modal neural structures or tokenizers. |
Simplify-Pro: A Two-level and Progressive LLM-based Framework for Auto Long Text Simplification (2026.findings-acl)
Copied to clipboard
| Challenge: | Existing studies have focused on lexical- and sentence-level simplification, leaving long text simplification comparatively unexplored . |
| Approach: | They propose a two-level and progressive LLM-based framework that establishes an effective paradigm for automatic long text simplification under diverse test scenarios. |
| Outcome: | The proposed framework outperforms advanced and proprietary LLMs in in-domain and out-of-domain simplification tasks and matches or outperformed existing LLM frameworks. |
Context-Sensitive Generation of Open-Domain Conversational Responses (C18-1)
Copied to clipboard
| Challenge: | Existing studies on single-turn conversation generation focus on coherence and context-sensitive generation of open-domain conversational responses. |
| Approach: | They propose static and dynamic attention based approaches for context-sensitive generation of open-domain conversational responses. |
| Outcome: | The proposed model outperforms all baselines on automatic and human evaluation on two public datasets. |
SLIP: Soft Label Mechanism and Key-Extraction-Guided CoT-based Defense Against Instruction Backdoor in APIs (2026.findings-acl)
Copied to clipboard
| Challenge: | Existing black-box instruction backdoors can detect poisoned inputs, but fail to recover correct outputs once the backdoor is activated. |
| Approach: | They propose a soft label mechanism and key-extraction-guided CoT-based defense against instruction backdoors in APIs (SLIP) they propose KCOT-based model to extract task-relevant keywords and phrases rather than only considering the single trigger or overall text semantics. |
| Outcome: | The proposed model reduces the average attack success rate to 25.13% and improves clean accuracy to 87.15% and outperforms state-of-the-art black-box defenses. |
Struc-Bench: Are Large Language Models Good at Generating Complex Structured Tabular Data? (2024.naacl-short)
Copied to clipboard
| Challenge: | Large Language Models (LLMs) have advanced capabilities but produce complex structured data. |
| Approach: | They propose a structure-aware fine-tuning method to bolster LLMs' performance by crafting format-specific instructions from the intended outputs. |
| Outcome: | The proposed method outperforms LLMs on all three formats and spans text tables, HTML, and LaTeX formats. |
Can LLMs Act as Historians? Evaluating Historical Research Capabilities of LLMs via the Chinese Imperial Examination (2026.acl-long)
Copied to clipboard
Lirong Gao, Zeqing Wang, Yuyan Cai, Jiayi Deng, Yanmei Gu, Yiming Zhang, Jia Zhou, Yanfei Zhang, Junbo Zhao
| Challenge: | Existing benchmarks assess basic knowledge breadth or lexical understanding, failing to capture higher-order skills that are central to historical research. |
| Approach: | They propose a benchmark anchored in the Chinese Imperial Examination system that assesses historical knowledge and lexical understanding. |
| Outcome: | The new benchmark aims to assess the ability of LLMs to process historical materials and documents. |
COIG-P: A High-Quality and Large-Scale Chinese Preference Dataset for Alignment with Human Values (2026.findings-eacl)
Copied to clipboard
Siwei Wu, JinCheng Ren, Xeron Du, Shuyue Guo, Xingwei Qu, Yiming Liang, Jie Liu, Yunwen Li, Tyler Loakman, Tianyu Zheng, Boyu Feng, Huaqing Yuan, Zili Wang, Jiaheng Liu, Wenhao Huang, Chenglin Cai, Haoran Que, Jian Yang, Yuelin Bai, Zekun Moore Wang, Zhouliang Yu, Qunshu Lin, Ding Pan, Yuchen Eleanor Jiang, Tiannan Wang, Wangchunshu Zhou, Shenzhi Wang, Xingyuan Bu, Minghao Liu, Guoyin Wang, Ge Zhang, Chenghua Lin
| Challenge: | Existing Chinese preference datasets suffer from limited scale, restricted domain coverage, and insufficiently rigorous data validation. |
| Approach: | They propose an LLM-based data annotation pipeline with no human intervention to annotate Chinese preference datasets. |
| Outcome: | The proposed pipeline outperforms existing Chinese preference datasets on AlignBench and Chinese Reward Benchmark. |
Kill two birds with one stone: generalized and robust AI-generated text detection via dynamic perturbations (2025.naacl-long)
Copied to clipboard
| Challenge: | Existing methods focus on model generalization or focus on robustness. |
| Approach: | They propose a model-based AIGT detection method that can be generalized and robust under two adversarial attacks. |
| Outcome: | The proposed method outperforms state-of-the-art methods for generalization and robustness under two text adversarial attacks. |
D-NET: A Pre-Training and Fine-Tuning Framework for Improving the Generalization of Machine Reading Comprehension (D19-58)
Copied to clipboard
Hongyu Li, Xiyuan Zhang, Yibing Liu, Yiming Zhang, Quan Wang, Xiangyang Zhou, Jing Liu, Hua Wu, Haifeng Wang
| Challenge: | MRC requires machines to understand text and answer questions about the text. |
| Approach: | They propose a simple system Baidu submitted for MRQA 2019 Shared Task that focused on generalization of machine reading comprehension (MRC) models. |
| Outcome: | The proposed system is ranked at top 1 of all participants in terms of averaged F1 score. |
R-Judge: Benchmarking Safety Risk Awareness for LLM Agents (2024.findings-emnlp)
Copied to clipboard
Tongxin Yuan, Zhiwei He, Lingzhong Dong, Yiming Wang, Ruijie Zhao, Tian Xia, Lizhen Xu, Binglin Zhou, Fangqi Li, Zhuosheng Zhang, Rui Wang, Gongshen Liu
| Challenge: | Large language models (LLMs) have shown compelling abilities in reasoning, decision-making, and instruction following. |
| Approach: | They propose a benchmark to evaluate the proficiency of large language models (LLMs) in judging and identifying safety risks given agent interaction records. |
| Outcome: | The proposed model outperforms the best-performing model, GPT-4o, while no other models significantly exceed the random. |
Reverse Constitutional AI: A Framework for Controllable Toxic Data Generation via Probability-Clamped RLAIF (2026.findings-acl)
Copied to clipboard
| Challenge: | Existing approaches to red teaming focus on searching for individual adversarial inputs. |
| Approach: | They propose a framework for automated adversarial data generation that inverts harmless constitution into constitution of toxicity and iteratively refining model outputs through critique–revision pipeline. |
| Outcome: | The proposed framework generates diverse, high-quality toxic data without human annotation and significantly improves semantic coherence without sacrificing adversarial strength. |
Learning to Ignore Adversarial Attacks (2023.eacl-main)
Copied to clipboard
| Challenge: | Despite the strong performance of current NLP models, they can be brittle against adversarial inputs. |
| Approach: | They propose a rationale model that explicitly learns to ignore adversarial tokens . their approach leads to sizable improvements in robustness over baseline models . |
| Outcome: | The proposed model outperforms data augmentation with adversarial examples and closes the gap between model performance and an attacked test set. |
Evo-Attacker: Memory-Augmented Reinforcement Learning for Long-Horizon Tool Attacks on LLM-MAS (2026.acl-long)
Copied to clipboard
| Challenge: | Existing tool attacks are limited by domain specificity or fixed and static templates. |
| Approach: | They propose an attack-based memory-augmented reinforcement learning process that constructs a dynamic attack memory and employs deliberative reasoning to retrieve adversarial patterns. |
| Outcome: | Evo-Attacker outperforms baselines in the long-horizon credit assignment challenge. |
Inhibitory Attacks on Backdoor-based Fingerprinting for Large Language Models (2026.acl-long)
Copied to clipboard
| Challenge: | Backdoor-based LLM fingerprinting is a promising solution for intellectual property protection . however, the vulnerability of existing LLMs for the ensemble scenario is unexplored . |
| Approach: | They propose two new fingerprinting attack methods to assess the robustness of LLM fingerprinting by token filter attack and sentence verification attack. |
| Outcome: | The proposed methods inhibit the fingerprint response while maintaining ensemble performance. |
Language Models of Code are Few-Shot Commonsense Learners (2022.emnlp-main)
Copied to clipboard
| Challenge: | Existing approaches to generate graphs using pre-trained language models hinder their ability to generate them correctly. |
| Approach: | They propose to frame structured commonsense reasoning tasks as code generation tasks instead of serializing the output graph as a flat list of nodes and edges. |
| Outcome: | The proposed approach outperforms natural-language LMs in three natural language tasks even when the downstream task does not involve source code at all. |
COIG-CQIA: Quality is All You Need for Chinese Instruction Fine-tuning (2025.findings-naacl)
Copied to clipboard
Yuelin Bai, Xeron Du, Yiming Liang, Leo Jin, Junting Zhou, Ziqiang Liu, Feiteng Fang, Mingshan Chang, Tianyu Zheng, Xincheng Zhang, Nuo Ma, Zekun Moore Wang, Ruibin Yuan, Haihong Wu, Hongquan Lin, Wenhao Huang, Jiajun Zhang, Chenghua Lin, Jie Fu, Min Yang, Shiwen Ni, Ge Zhang
| Challenge: | Existing datasets for Chinese instruction tuning are not well-aligned with Chinese users’ interaction patterns. |
| Approach: | They propose to use Chinese instruction tuning datasets to improve instruction fine-tuning for Chinese users. |
| Outcome: | The proposed dataset shows that Chinese models achieve competitive performance in diverse benchmarks. |