Papers by Xiangzheng Zhang
From Detection to Understanding: Multi-Turn Reasoning for Video Misinformation Analysis (2026.acl-long)
Copied to clipboard
| Challenge: | Existing benchmarks focus on binary veracity judgments and do not evaluate process-level justifications for misinformation models. |
| Approach: | They propose a video misinformation analysis benchmark that assesses reasoning in video misinterpretation. |
| Outcome: | The proposed framework improves reasoning accuracy and explanation quality compared to existing models . it covers 12 fine-grained deception categories and progresses from perceptual attribution to intent and persuasion analysis. |
Reasoning-Augmented Conversation for Multi-Turn Jailbreak Attacks on Large Language Models (2025.findings-emnlp)
Copied to clipboard
Zonghao Ying, Deyue Zhang, Zonglei Jing, Yisong Xiao, Quanchen Zou, Aishan Liu, Siyuan Liang, Xiangzheng Zhang, Xianglong Liu, Dacheng Tao
| Challenge: | Existing methods for implementing multi-turn jailbreaks struggle to balance semantic coherence with attack effectiveness, resulting in benign semantic drift or ineffective detection evasion. |
| Approach: | They propose a framework that reformulates harmful queries into benign reasoning tasks and leverages LLMs’ strong reasoning capabilities to compromise safety alignment. |
| Outcome: | The proposed framework achieves state-of-the-art attack effectiveness in complex conversational scenarios, with average ASRs increasing by up to 96%. |
Light-R1: Curriculum SFT, DPO and RL for Long COT from Scratch and Beyond (2025.acl-industry)
Copied to clipboard
Liang Wen, Yunke Cai, Fenrui Xiao, Xin He, Qi An, Zhenyu Duan, Yimin Du, Junchen Liu, Tanglifu Tanglifu, Xiaowei Lv, Haosheng Zou, Yongchao Deng, Shousheng Jia, Xiangzheng Zhang
| Challenge: | Experimental results show that opensource curriculum training is more effective when distinct datasets are available for different training stages. |
| Approach: | They propose an opensource suite for training long reasoning models using publicdata and models. |
| Outcome: | The proposed model outperforms DeepSeek-R1-DistillQwen-32B models in math reasoning. |
Thinking with Reasoning Skills: Fewer Tokens, More Accuracy (2026.acl-industry)
Copied to clipboard
| Challenge: | Reasoning LLMs often spend tokens on long intermediate reasoning traces when solving new problems. |
| Approach: | They propose to store reusable reasoning skills distilled from extensive deliberation and trial-and-error exploration and retrieve these skills at inference time to guide future reasoning. |
| Outcome: | The proposed approach reduces reasoning tokens while improving overall performance on coding and mathematical reasoning tasks. |
TriPlay-RL: Tri-Role Self-Play Reinforcement Learning for LLM Safety Alignment (2026.acl-long)
Copied to clipboard
Zhewen Tan, Wenhan Yu, Jianfeng Si, Tongxin Liu, Kaiqi Guan, Huiyan Jin, Jiawen Tao, Xiaokun Yuan, Xiangzheng Zhang, Duohe Ma, Tong Yang, Lin Sun
| Challenge: | Existing approaches to safety alignment of large language models rely on costly manual annotations or human review. |
| Approach: | They propose a closed-loop reinforcement learning framework called TriPlay-RL that enables iterative collaboration among three roles with near-zero manual annotation. |
| Outcome: | The proposed framework achieves 20%–50% improvement in adversarial effectiveness while preserving high output diversity while achieving 10%–30% gains in safety performance without degrading general reasoning capability. |
DMN: A Compositional Framework for Jailbreaking Multimodal LLMs with Multi-Image Inputs (2026.acl-long)
Copied to clipboard
| Challenge: | Existing jailbreak methods only use a single image, restricting the attack space . Existing frameworks only use single image to distribute harmful requests across multiple images . |
| Approach: | They propose a compositional jailbreak framework that leverages Distributed instruction, Multimodal evidence and a Number chain task to fully enhance the jailbreak performance. |
| Outcome: | The proposed framework achieves attack success rates of over 90% on GPT-4o, Gemini-2.5-pro and Claude Sonnet 4 . |
Large Language Models Badly Generalize across Option Length, Problem Types, and Irrelevant Noun Replacements (2025.emnlp-main)
Copied to clipboard
| Challenge: | Existing benchmarks have exposed patterns and may not truly assess generalization ability of Large Language Models (LLMs). |
| Approach: | They propose a “Generalization Stress Test” to assess Large Language Models’ generalization ability under slight and controlled perturbations, including option length, problem types, and irrelevant noun replacements. |
| Outcome: | The proposed test shows that LLMs exhibit severe accuracy drops and unexpected biases when faced with minor but content-preserving modifications. |
When Good OCR Is Not Enough: Benchmarking OCR Robustness for Retrieval-Augmented Generation (2026.acl-industry)
Copied to clipboard
Lin Sun, null Wangdexian, Jingang Huang, Linglin Zhang, Change Jia, Zhengwei Cheng, Xiangzheng Zhang
| Challenge: | Existing OCR benchmarks rely on character-level metrics to measure downstream performance . high OCR accuracy does not translate into strong downstream performance, authors say . |
| Approach: | They propose an OCR benchmark for industrial RAG systems that measures character-level metrics . they find that high OCR accuracy does not translate into strong downstream RAG performance . |
| Outcome: | The proposed benchmark shows that high OCR accuracy does not translate into strong downstream performance . structural and semantic errors can cause substantial retrieval failures even when WER/CER remains low. |
Chain-of-Thought Matters: Improving Long-Context Language Models with Reasoning Path Supervision (2025.findings-emnlp)
Copied to clipboard
Dawei Zhu, Xiyu Wei, Guangxiang Zhao, Wenhao Wu, Haosheng Zou, Junfeng Ran, null XWang, Lin Sun, Xiangzheng Zhang, Sujian Li
| Challenge: | Recent advances in Large Language Models (LLMs) have highlighted the challenge of handling long-context tasks. |
| Approach: | They propose a chain-of-thought framework that teaches models to generate high-quality reasoning paths for enhanced long-context performance. |
| Outcome: | The proposed framework generalizes across most long-context scenarios and amplifys with increasing context length. |