Changsheng Wang

PhD student, CSE, Michigan State University

Joined

September 2022

Names

Changsheng Wang (Preferred)

Emails

****@gmail.com (Confirmed)

****@mail.ustc.edu.cn (Confirmed)

****@msu.edu (Confirmed)

Personal Links

Career & Education History

PhD student

CSE, Michigan State University (msu.edu)

2024 – Present

Advisors, Relations & Conflicts

PhD Advisor

Sijia Liu

2024 – Present

Expertise

Vision Unlearning

2024 – 2025

Publications

Downgrade to Upgrade: Optimizer Simplification Enhances Robustness in LLM Unlearning
Yicheng Lang, Yihua Zhang, Chongyu Fan, Changsheng Wang, Jinghan Jia, Sijia Liu
- ICLR 2026 Poster
- Readers: Everyone
LLM Unlearning Under the Microscope: A Full-Stack View on Methods and Metrics
Chongyu Fan, Changsheng Wang, Yancheng Huang, Soumyadeep Pal, Sijia Liu
- Submitted to ICLR 2026
- Readers: Everyone
Reasoning Model Unlearning: Forgetting Traces, Not Just Answers, While Preserving Reasoning Skills
Changsheng Wang, Chongyu Fan, Yihua Zhang, Jinghan Jia, Dennis Wei, Parikshit Ram, Nathalie Baracaldo, Sijia Liu
- CoRR 2025
- Readers: Everyone
Invariance Makes LLM Unlearning Resilient Even to Unanticipated Downstream Fine-Tuning
Changsheng Wang, Yihua Zhang, Jinghan Jia, Parikshit Ram, Dennis Wei, Yuguang Yao, Soumyadeep Pal, Nathalie Baracaldo, Sijia Liu
- CoRR 2025
- Readers: Everyone
Noisy But Forgotten: LLM Unlearning are Robust against Perturbed Data in the Wild
Changsheng Wang, Yihua Zhang, Jinghan Jia, Dennis Wei, Sijia Liu
- MUGen @ ICML 2025 Poster
- Readers: Everyone
Rethinking Unlearning for Large Reasoning Models
Changsheng Wang, Chongyu Fan, Yihua Zhang, Jinghan Jia, Dennis Wei, Parikshit Ram, Nathalie Baracaldo, Sijia Liu
- MUGen @ ICML 2025 Poster
- Readers: Everyone
The Fragile Truth of Saliency: Improving LLM Input Attribution via Attention Bias Optimization
Yihua Zhang, Changsheng Wang, Yiwei Chen, Chongyu Fan, Jinghan Jia, Sijia Liu
- NeurIPS 2025 spotlight
- Readers: Everyone
LLM Unlearning Reveals a Stronger-Than-Expected Coreset Effect in Current Benchmarks
Soumyadeep Pal, Changsheng Wang, James Diffenderfer, Bhavya Kailkhura, Sijia Liu
- CoRR 2025
- Readers: Everyone
LLM Unlearning Reveals a Stronger-Than-Expected Coreset Effect in Current Benchmarks
Soumyadeep Pal, Changsheng Wang, James Diffenderfer, Bhavya Kailkhura, Sijia Liu
- COLM 2025
- Readers: Everyone
Invariance Makes LLM Unlearning Resilient Even to Unanticipated Downstream Fine-Tuning
Changsheng Wang, Yihua Zhang, Jinghan Jia, Parikshit Ram, Dennis Wei, Yuguang Yao, Soumyadeep Pal, Nathalie Baracaldo, Sijia Liu
- ICML 2025 poster
- Readers: Everyone

View all 13 publications