Research

I am interested in answering questions in Trustworthy AI, Differential privacy, Uncertainty Quantification, and Federated Learning, for which I use theoretical tools from statistics and optimization, complemented by rigorous experimentation. More concretely, in the near future, I am interested in the following directions:

1. Studying the empirical privacy leakage for modern ML models (Foundation Models/LLMs) in realistic attack scenarios and moving beyond DP to find application-relevant definitions to evaluate models on privacy, robustness, fairness and copyright.
2. Developing mitigation strategies and training algorithms that ensure foundation models adhere to trustworthy behavior.
3. Building better (trustworthy) algorithms and systems for practically relevant ML and data analytics tasks like recommendations, ranking, frequency estimation, etc.
4. Leveraging the capabilities of foundation models for private synthetic data generation to help with private training in low resource tasks.

Publications

Check Google Scholar for an up to date list.

• Auditing Private Prediction
  K. Chadha, M. Jagielski, C. Choquette-Choo, M. Nasr, and N. Papernot
  

• Resampling methods for private statistical inference
  K. Chadha, J. C. Duchi and R. Kuditipudi
  

• Differentially Private Heavy Hitter Detection using Federated Analytics
  K. Chadha J. Chen, J. C. Duchi, V. Feldman, H. Hashemi, O. Javidbakht, A. McMillan, and K. Talwar
  Workshops: Federated Learning and Analytics in Practice, TPDP

• Federated Asymptotics: A model for evaluating federated learning algorithms
  K. Chadha, G. Cheng, and J. C. Duchi
  AISTATS 23

• Private optimization in the interpolation regime: faster rates and hardness results
  K. Chadha, H. Asi, G. Cheng*, and J. C. Duchi
  ICML 22

• Accelerated, optimal, and parallel: Some results on model-based stochastic optimization
  K. Chadha, G. Cheng, and J. C. Duchi
  ICML 22

• Minibatch stochastic approximate proximal point methods
  K. Chadha, H. Asi, G. Cheng*, and J. C. Duchi
  NeurIPS 2020

• Efficiency fairness tradeoff in battery sharing
  K. Chadha, A. A. Kulkarni and J. Nair
  Operations Research Letters, 2021

• Aggregate play and welfare in strategic interactions on networks
  K. Chadha and A. A. Kulkarni
  Journal of Mathematical Economics, 2020

• On independent cliques and linear complementarity problems
  K. Chadha and A. A. Kulkarni
  IJPAM, 2022

• A reinforcement learning algorithm for restless bandits
  V. Borkar and K. Chadha
  Indian Control Conference, 2018