Ayush Sekhari

Machine Unlearning: I have developed foundational techniques and efficient methods for securely removing data from ML models post-deployment, ensuring privacy and compliance with evolving data regulations and operational needs.

• Remember What You Want to Forget: Algorithms for Machine Unlearning,
  with Jayadev Acharya, Gautam Kamath, and Ananda Theertha Suresh (NeurIPS 2021).

• Ticketed Learning-Unlearning Schemes,
  with Badih Ghazi, Pritish Kamath, Ravi Kumar, Pasin Manurangsi, and Chiyuan Zhang (COLT 2023).

Interactive Learning: I have created theoretically rigorous and practical algorithms for learning in dynamic, real-world environments. This includes new algorithmic frameworks for Reinforcement Learning (RL) and imitation learning.

• Selective Sampling and Imitation Learning via Online Regression,
  with Karthik Sridharan, Wen Sun, and Runzhe Wu (NeurIPS 2023).

• On the Complexity of Adversarial Decision Making,
  with Dylan J. Foster, Alexander Rakhlin and Karthik Sridharan. (NeurIPS 2022, oral presentation).

• Hybrid RL: Using Both Offline and Online Data Can Make RL Efficient,
  with Yuda Song, Yifei Zhou, J. Andrew Bagnell, Akshay Krishnamurthy, Wen Sun (ICLR 2023).

Stochastic Optimization: I have advanced our understanding of first-order methods and developed new optimal algorithms for learning in large-scale, complex, non-convex models that are central to modern ML.

• SGD: The role of Implicit Regularization, Batch-size and Multiple Epochs,
  with Satyen Kale and Karthik Sridharan (NeurIPS 2021).

• From Gradient Flow on Population Loss to Learning with Stochastic Gradient Descent,
  with Satyen Kale, Jason D. Lee, Chris De Sa and Karthik Sridharan (NeurIPS 2022).

• The Complexity of Making the Gradient Small in Stochastic Convex Optimization,
  with Dylan Foster, Ohad Shamir, Nathan Srebro, Karthik Sridharan and Blake Woodworth (COLT 2019, Best Student Paper Award).

---

Recent News

• [Apr'24] New paper on arXiv on efficient Agnostic RL under reset access to the environment. Accepted to COLT 2025.
• [Jan'24] New paper on arXiv on practical watermarking of Large Language Models (LLMs). Accepted to ICML 2025.
• [Jan'24] Our paper on deterministic linear bellman complete MDPs accepted at ICLR 2025 as Oral Presentation.
• [Jan'24] Our paper on evaluating machine unlearning algorithms accepted at ICLR 2025.
• [Dec'24] I am attending NeurIPS 2024. Let us meet if you are around?
• [Dec'24] Giving a short talk at the Princeton Language and Intelligence seminar on "Machine Unlearning: Where does the practice stand?".
• [Dec'24] Giving a talk at Rutgers University on learning-unlearning schemes.
• I will be an area chair for Algorithmic Learning Theory (ALT) 2025.

• Yunbei Xu and I are organizing a session on Theoretical Aspects of Interactive Learning under Applied Probability Society at INFORMS 2024 from Oct 20-23. Our awesome set of speakers include Noah Golowich, Zeyu Jia, Kevin Jamieson, and Yunzhong Xu.
• I will be attending Cornell's ORIE Young Researchers Workshop in Ithaca, NY from Oct 9-11. Let us chat if you are around!
• I will be attending Mathematical and Scientific Foundations of Deep Learning Annual Meeting at Simons Institute, NYC from Sept 26-27. Let us chat if you are around!
• Two papers accepted at ICLR 2024. Looking forward to the conference in Vienna.