Publications

2025

1. Unlearning Isn’t Invisible: Detecting Unlearning Traces in LLMs from Model Outputs

   Soumyadeep Pal*, Yiwei Chen*, Yimeng Zhang, Qing Qu, and Sijia Liu

   Oral at Machine Unlearning for Generative AI workshop 2025

   Paper Video

   @article{chen2025unlearning,
     title = {Unlearning Isn't Invisible: Detecting Unlearning Traces in LLMs from Model Outputs},
     author = {Pal, Soumyadeep and Chen, Yiwei and Zhang, Yimeng and Qu, Qing and Liu, Sijia},
     journal = {Oral at Machine Unlearning for Generative AI workshop},
     year = {2025}
   }

2. LLM Unlearning Reveals a Stronger-Than-Expected Coreset Effect in Current Benchmarks

   Soumyadeep Pal*, Changsheng Wang*, James Diffenderfer, Bhavya Kailkhura, and Sijia Liu

   In Second Conference on Language Modeling (COLM) 2025

   Paper Code

   @inproceedings{pal2025llm,
     title = {LLM Unlearning Reveals a Stronger-Than-Expected Coreset Effect in Current Benchmarks},
     author = {Pal, Soumyadeep and Wang, Changsheng and Diffenderfer, James and Kailkhura, Bhavya and Liu, Sijia},
     booktitle = {Second Conference on Language Modeling (COLM)},
     year = {2025},
   }

3. 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, and Sijia Liu

   In Forty-second International Conference on Machine Learning (ICML) 2025

   Paper Code

2024

1. Backdoor Secrets Unveiled: Identifying Backdoor Data with Optimized Scaled Prediction Consistency

   Soumyadeep Pal, Yuguang Yao, Ren Wang, Bingquan Shen, and Sijia Liu

   In The Twelfth International Conference on Learning Representations (ICLR) 2024

   Paper Code

2023

1. Towards understanding how self-training tolerates data backdoor poisoning

   Soumyadeep Pal, Ren Wang, Yuguang Yao, and Sijia Liu

   Best paper finalist @ The AAAI’s Workshop on Artificial Intelligence Safety 2023

   Paper