Papers by Jinghan Jia
SOUL: Unlocking the Power of Second-Order Optimization for LLM Unlearning (2024.emnlp-main)
Copied to clipboard
Jinghan Jia, Yihua Zhang, Yimeng Zhang, Jiancheng Liu, Bharat Runwal, James Diffenderfer, Bhavya Kailkhura, Sijia Liu
| Challenge: | Large Language Models (LLMs) have highlighted the need for effective unlearning mechanisms to comply with data regulations and ethical AI practices. |
| Approach: | They propose a second-order optimization-based LLM unlearning framework which extends the static, one-shot model update using influence unlearning to a dynamic, iterative unlearning process. |
| Outcome: | The proposed framework outperforms first-order methods across unlearning tasks, models, and metrics. |
I2E: From Image Pixels to Actionable Interactive Environments for Text-Guided Image Editing (2026.acl-long)
Copied to clipboard
Jinghan Yu, Junhao Xiao, Chenyu Zhu, Jiaming Li, Jia Li, HanMing Deng, Xirui Wang, Guoli Jia, Jianjun Li, Xiang Bai, Bowen Zhou, Zhiyuan Ma
| Challenge: | Existing text-guided image editing methods rely on end-to-end pixel-level inpainting paradigm . existing models lack such intermediate representations and Reasoning-then-action process . |
| Approach: | They propose a "Decompose-then-Action" paradigm that revisits image editing as an actionable interaction process within a structured environment. |
| Outcome: | The proposed paradigm outperforms existing methods in compositional editing tasks. |
SEUF: Is Unlearning One Expert Enough for Mixture-of-Experts LLMs? (2025.acl-long)
Copied to clipboard
| Challenge: | Recent advances in LLMs unlearning have shown remarkable success in removing unwanted data-model influences while preserving the model’s utility for legitimate knowledge. |
| Approach: | They propose a Selected-Expert Unlearning Framework (SEUF) that combines expert attribution and an anchor loss to ensure controlled unlearning. |
| Outcome: | Experiments show that the proposed framework improves forget quality and model utility by 35% on MoE LLMs across benchmarks and LLM architectures compared to standard unlearning algorithms . |
Reasoning Model Unlearning: Forgetting Traces, Not Just Answers, While Preserving Reasoning Skills (2025.emnlp-main)
Copied to clipboard
Changsheng Wang, Chongyu Fan, Yihua Zhang, Jinghan Jia, Dennis Wei, Parikshit Ram, Nathalie Baracaldo, Sijia Liu
| Challenge: | Existing methods for LRM unlearning overlook critical information leakage in reasoning traces, even when final answers are successfully removed. |
| Approach: | They propose a method that suppresses reasoning traces while preserving the model's general reasoning ability. |
| Outcome: | The proposed method significantly reduces reasoning trace leakage and achieves strong performance across reasoning and safety benchmarks, including WMDP, StrongReject, JBB-Behaviors and WildJailbreak. |
BLUR: A Bi-Level Optimization Approach for LLM Unlearning (2026.eacl-long)
Copied to clipboard
Hadi Reisizadeh, Jinghan Jia, Zhiqi Bu, Bhanukiran Vinzamuri, Anil Ramakrishna, Kai-Wei Chang, Volkan Cevher, Sijia Liu, Mingyi Hong
| Challenge: | Existing algorithms to unlearn knowledge and capabilities from large datasets are unclear how to best formulate the unlearning problem. |
| Approach: | They propose to model the hierarchical structure of the unlearning problem, where the forget problem takes priority over the retain problem, and propose an algorithm that aims to unlearn knowledge and capabilities. |
| Outcome: | The proposed algorithm outperforms all state-of-the-art algorithms across unlearning tasks, models, and metrics. |
Cracking the Code of Hallucination in LVLMs with Vision-aware Head Divergence (2025.acl-long)
Copied to clipboard
Jinghan He, Kuan Zhu, Haiyun Guo, Junfeng Fang, Zhenglin Hua, Yuheng Jia, Ming Tang, Tat-Seng Chua, Jinqiao Wang
| Challenge: | Existing methods focus on alignment training or decoding refinements but address symptoms at the generation stage without probing the underlying causes. |
| Approach: | They propose a training-free approach to mitigate hallucination by enhancing the role of vision-aware attention heads. |
| Outcome: | The proposed method achieves superior performance compared to state-of-the-art approaches in mitigating hallucinations while maintaining high efficiency with negligible additional time overhead. |
Steering LVLMs via Sparse Autoencoder for Hallucination Mitigation (2025.findings-emnlp)
Copied to clipboard
| Challenge: | Existing approaches to address hallucinations in large vision-language models require substantial computational cost and time. |
| Approach: | They propose to leverage sparse autoencoders to identify semantic directions closely associated with faithfulness or hallucination, extracting more precise and disentangled hallucinian-related representations. |
| Outcome: | The proposed method outperforms existing decoding approaches while maintaining transferability across different model architectures with negligible additional time overhead. |