Papers by Zhaoyang Wang
Evaluating the Validity of Word-level Adversarial Attacks with Large Language Models (2024.findings-acl)
Copied to clipboard
| Challenge: | Existing adversarial examples can generate invalid adversarials due to significant changes in semantic meanings compared to their originals. |
| Approach: | They propose to use a large language model to evaluate adversarial examples by semantic constraints. |
| Outcome: | The proposed method can generate valid adversarial examples even when they are not equipped with semantic constraints. |
Democratizing Reasoning Ability: Tailored Learning from Large Language Model (2023.emnlp-main)
Copied to clipboard
Zhaoyang Wang, Shaohan Huang, Yuxuan Liu, Jiahai Wang, Minghui Song, Zihan Zhang, Haizhen Huang, Furu Wei, Weiwei Deng, Feng Sun, Qi Zhang
| Challenge: | Large language models (LLMs) exhibit impressive emergent abilities in natural language processing, but their democratization is hindered due to huge computation requirements and closed-source nature. |
| Approach: | They propose a tailored learning approach to distill the exclusive reasoning ability to smaller LMs to facilitate democratization. |
| Outcome: | The proposed approach enables the democratization of the exclusive reasoning ability by leveraging the black-box model as a reasoning teacher. |
UECA-Prompt: Universal Prompt for Emotion Cause Analysis (2022.coling-1)
Copied to clipboard
| Challenge: | Existing methods adopt fine-tuning paradigm to solve certain types of ECA tasks. Existing models suffer from dataset bias. |
| Approach: | They propose a universal prompt tuning method to solve different ECA tasks in a unified framework and a sequential learning module to ease the dataset bias. |
| Outcome: | The proposed method achieves competitive performance on the ECA datasets. |
TESTEVAL: Benchmarking Large Language Models for Test Case Generation (2025.findings-naacl)
Copied to clipboard
Wenhan Wang, Chenyuan Yang, Zhijie Wang, Yuheng Huang, Zhaoyang Chu, Da Song, Lingming Zhang, An Ran Chen, Lei Ma
| Challenge: | Existing methods to generate test cases using large language models are limited in their ability to generate unit test cases. |
| Approach: | They propose a test case generation benchmark that uses large language models to generate unit test cases. |
| Outcome: | The proposed test case generation benchmarks compare LLMs with commercial and open-source LLM platforms and find that they lack the ability to comprehend program logic and execution paths. |
RMLM: A Flexible Defense Framework for Proactively Mitigating Word-level Adversarial Attacks (2023.acl-long)
Copied to clipboard
| Challenge: | Existing defenses focus on improving robustness of the victim model in training, but neglect to mitigate adversarial attacks during inference. |
| Approach: | They propose a framework that confuses attackers and corrects adversarial contexts . their framework helps improve the robustness of the victim model during inference . |
| Outcome: | The proposed framework improves the robustness of the victim model in training . it also corrects abnormal contexts in the representation level and filtering out examples . |
MMAU: A Holistic Benchmark of Agent Capabilities Across Diverse Domains (2025.findings-naacl)
Copied to clipboard
Guoli Yin, Haoping Bai, Shuang Ma, Feng Nan, Yanchao Sun, Zhaoyang Xu, Shen Ma, Jiarui Lu, Xiang Kong, Aonan Zhang, Dian Ang Yap, Yizhe Zhang, Karsten Ahnert, Vik Kamath, Mathias Berglund, Dominic Walsh, Tobias Gindele, Juergen Wiest, Zhengfeng Lai, Xiaoming Simon Wang, Jiulong Shan, Meng Cao, Ruoming Pang, Zirui Wang
| Challenge: | Existing benchmarks focus on specific application scenarios, emphasizing task completion but failing to dissect the underlying skills that drive these outcomes. |
| Approach: | They propose a Massive Multitask Agent Understanding benchmark that evaluates LLMs across five domains and offline tasks. |
| Outcome: | The Massive Multitask Agent Understanding (MMAU) benchmark evaluates models across five domains including Tool-use, Directed Acyclic Graph (DAG) QA, Data Science and Machine Learning coding, Contest-level programming and Mathematics. |
OS-Symphony: A Holistic Framework for Robust and Generalist Computer-Using Agents (2026.acl-long)
Copied to clipboard
Bowen Yang, Kaiming Jin, Zhenyu Wu, Zhaoyang Liu, Qiushi Sun, Zehao Li, JingJing Xie, Zhoumianze Liu, Fangzhi Xu, Kanzhi Cheng, Yian Wang, Qingyun Li, Yu Qiao, Zun Wang, Zichen Ding
| Challenge: | Vision-Language Models (VLMs) lack visual-aware tutorial retrieval and historical visual context curation and pruning. |
| Approach: | They propose a framework that integrates an orchestrator and a Reflection-Memory Agent for robust automation. |
| Outcome: | Experimental results show that OS-Symphony delivers substantial performance gains across model scales. |
All Changes May Have Invariant Principles: Improving Ever-Shifting Harmful Meme Detection via Design Concept Reproduction (2026.acl-long)
Copied to clipboard
Ziyou Jiang, Mingyang Li, Junjie Wang, Yuekai Huang, Jie Huang, Zhiyuan Chang, Zhaoyang Li, Qing Wang
| Challenge: | Existing methods for harmful meme detection only learn the combination of harmful elements and lack understanding of these implicit expressions. |
| Approach: | They propose a method that detects harmful memes by replicating the design concept of malicious users. |
| Outcome: | The proposed method achieves the highest accuracy with 81.1% and has slight accuracy decreases when generalized to type-shifting and temporal-evolving memes. |
Know Thy Enemy: Securing LLMs Against Prompt Injection via Diverse Data Synthesis and Instruction-Level Chain-of-Thought Learning (2026.findings-acl)
Copied to clipboard
Zhiyuan Chang, Mingyang Li, Yuekai Huang, Ziyou Jiang, Xiaojun Jia, Qian Xiong, Junjie Wang, Zhaoyang Li, Qing Wang
| Challenge: | Large language model (LLM)-integrated applications face security vulnerabilities from prompt injection (PI) attacks. |
| Approach: | They propose a model enhancement method that synthesizes diverse training data and employs instruction-level chain-of-thought fine-tuning to enable LLMs to effectively identify and reject malicious instructions regardless of their source or position in the context. |
| Outcome: | The proposed method outperforms baselines in three critical dimensions while maintaining utility performance without degradation. |
SynthAgent: Adapting Web Agents with Synthetic Supervision (2026.acl-long)
Copied to clipboard
Zhaoyang Wang, Yiming Liang, Xuchao Zhang, Qianhui Wu, Siwei Han, Anson Bastos, Rujia Wang, Chetan Bansal, Baolin Peng, Jianfeng Gao, Saravan Rajmohan, Huaxiu Yao
| Challenge: | Existing studies have focused on synthetic supervision but have encountered data quality issues. |
| Approach: | They propose a fully synthetic supervision framework that aims at improving data quality via dual refinement of both tasks and trajectories. |
| Outcome: | The proposed framework outperforms existing methods on standardized benchmarks and shows promising results on a standardized test. |
Wait, We Don’t Need to “Wait”! Removing Thinking Tokens Improves Reasoning Efficiency (2025.findings-emnlp)
Copied to clipboard
| Challenge: | Recent advances in large reasoning models often introduce significant overthinking . this leads to verbose and redundant outputs that hinder efficiency. |
| Approach: | They propose a plug-and-play solution that disables explicit self-reflection . it suppresses tokens such as "Wait" and "Hmm" during inference . |
| Outcome: | The proposed approach reduces chain-of-thought trajectory length by up to 27%–51% in five R1-style model series without compromising model utility. |
Verifiable Format Control for Large Language Model Generations (2025.findings-naacl)
Copied to clipboard
| Challenge: | Existing methods focus on benchmarking general instruction following while overlooking how to improve specific format following ability for small LLMs. |
| Approach: | They propose to synthesize massive datasets to improve LLMs' format following abilities by using a verifiable format following feature. |
| Outcome: | The proposed method improves the format following ability of small LLMs with about 7B parameters. |
Enhancing Visual-Language Modality Alignment in Large Vision Language Models via Self-Improvement (2025.findings-naacl)
Copied to clipboard
Xiyao Wang, Jiuhai Chen, Zhaoyang Wang, Yuhang Zhou, Yiyang Zhou, Huaxiu Yao, Tianyi Zhou, Tom Goldstein, Parminder Bhatia, Taha Kass-Hout, Furong Huang, Cao Xiao
| Challenge: | Existing methods for visual and language alignment depend on external models or data, leading to uncontrollable and unstable results. |
| Approach: | They propose a framework that enhances visual and language alignment without external dependencies by incorporating an in-context self-critic mechanism that constructs preference pairs for tuning. |
| Outcome: | The proposed framework outperforms existing methods and improves performance on 14 hallucination and comprehensive benchmarks. |
Synergistic Weak-Strong Collaboration by Aligning Preferences (2025.acl-long)
Copied to clipboard
Yizhu Jiao, Xuchao Zhang, Zhaoyang Wang, Yubo Ma, Zhun Deng, Rujia Wang, Chetan Bansal, Saravan Rajmohan, Jiawei Han, Huaxiu Yao
| Challenge: | Current Large Language Models excel in general reasoning yet struggle with specialized tasks requiring proprietary or domain-specific knowledge. |
| Approach: | They propose a collaborative framework that pairs a specialized weak model with a general strong model to optimize collaboration. |
| Outcome: | The proposed framework outperforms each model alone by leveraging complementary strengths. |