Papers

The research I’m most known for includes: the first SFT method for knowledge removal from models, the first algorithm with “provable” guarantees for learning an unbiased model with respect to exponentially many protected groups, the best known bounds connecting differential privacy and adaptive data analysis, and investigating the privacy properties of LLMs.

32. Humanity’s Sixth Sense: Benchmarking Intuitive Visual Reasoning in Multimodal Models [Preprint]

31. GOPHER: Optimization-based Phenotype Randomization for Genome-Wide Association Studies with Differential Privacy [RECOMB ’26] Best Paper Award! (single paper awarded each year)

30. Firm Foundations for Membership Inference in Language Models [Findings of the ACL ’26]

29. Machine Unlearning Doesn’t Do What You Think: Lessons for Generative AI Policy, Research, and Practice [NeurIPS ’25 Position Paper]

28. Efficient Verification of Data Attribution, [NeurIPS ’25]

27. Privacy Risks in Large Language Models: A Survey

26. Attribute-to-Delete: Machine Unlearning via Datamodel Matching, new paper using techniques from data attribution for unlearning! [ICLR ’25]

Machine Unlearning Fails to Remove Data Poisoning Attacks [ICLR ’25]

25. Private Regression in Multiple Outcomes (PRIMO) [TMLR ’25]

24. Better Counterfactual Model Reasoning with Submodular Quadratic Component Models [NeurIPS ’24 ATTRIB Workshop]

23. In-Context Unlearning: Language Models are Few Shot Unlearners [ICML ‘24]

22. Feature Importance Disparities for Dataset Bias Investigations [ICML ‘24] Code

21. MusCAT: A multi-scale, hybrid federated system for privacy-preserving epidemic surveillance and risk prediction [2nd Place Grand Prize Winner, 1st Place Phase 1 of the US/UK Privacy Challenge, ’23] White House Announcement

20. MoPe: Perturbation-based Privacy Attacks Against Language Models [EMNLP ’23, NEURIPS SoLaR Workshop ’23]

19. On the Privacy Risks of Algorithmic Recourse Code [AI STATS ’23]

18. Adaptive Machine Unlearning [NEURIPS ’21]

17. Data Privacy in Practice at LinkedIn [Harvard Business School Case Study ’22]

16. Descent-to-Delete: Gradient-Based Methods for Machine Unlearning [Algorithmic Learning Theory ’21] Code

15. Eliciting and Enforcing Subjective Individual Fairness [FORC ’21]

14. Optimal, Truthful, and Private Securities Lending [ACM AI in Finance ’20, NEURIPS Workshop on Robust AI in Financial Services ’19] selected for oral presentation!

13. Differentially Private Objective Perturbation: Beyond Smoothness and Convexity [ICML ’20, NEURIPS Workshop on Privacy in ML ’19]

12. A New Analysis of Differential Privacy’s Generalization Guarantees [ITCS ’20] regular talk slot!

11. The Role of Interactivity in Local Differential Privacy [FOCS ’19]

10. How to use Heuristics for Differential Privacy [FOCS ’19] Video.

9. An Empirical Study of Rich Subgroup Fairness for Machine Learning [ACM FAT* ’19, ML track]

Led development on package integrated into the IBM AI Fairness 360 package here. AIF360 development branch on my Github, with a stand-alone package developed by the AlgoWatch Team.

8. Fair Algorithms for Learning in Allocation Problems [ACM FAT* ’19, ML track]

7. Preventing Fairness Gerrymandering: Auditing and Learning for Subgroup Fairness [ICML ’18, EC MD4SG ’18]

6. Mitigating Bias in Adaptive Data Gathering via Differential Privacy [ICML ’18]

5. Accuracy First: Selecting a Differential Privacy Level for Accuracy Constrained ERM [NIPS ’17, Journal of Privacy and Confidentiality ’19]