Sham Kakade

Sham Machandranath Kakade
Alma mater	Caltech University College London
Scientific career
Fields	Computer Science, Artificial Intelligence
Institutions	Toyota Technological Institute at Chicago Wharton Microsoft Research University of Washington Harvard University
Doctoral advisor	Peter Dayan

Sham Machandranath Kakade is an American computer scientist. He is a Gordon McKay Professor in Computer Science at Harvard University, with a joint appointment in the Department of Statistics. Kakade is a co-director of the Kempner Institute for the Study of Natural and Artificial Intelligence. He co-founded the Algorithmic Foundations of Data Science Institute.

Education and Career

Kakade earned a Bachelor of Science in Physics from the California Institute of Technology and a PhD from the Gatsby Computational Neuroscience Unit at University College London, under the supervision of Peter Dayan. Prior to his current position at Harvard, he served as a Principal Researcher at Microsoft Research, an assistant professor at the Toyota Technological Institute at Chicago and Wharton, and a professor at the University of Washington.

Research

Kakade's research includes work on Reinforcement Learning, Tensor-Algebraic methods, and Convex optimization.

Reinforcement Learning

Kakade's doctoral work helped established statistical frameworks used in the study of sample complexity in reinforcement learning. He co-developed methods in policy optimization, including early work on natural policy gradient, conservative policy iteration. Kakade has contributed to theoretical analyses of reinforcement learning algorithms with provable performance guarantees.

Bandit Models

Kakade has worked extensively on multi-armed and structured bandit models, including linear and Gaussian process-based bandit. He co-authored "Gaussian Process Optimization in the Bandit Setting: No Regret and Experimental Design," which studied Gaussian process methods in a nonparametric bandit setting. The work established regret bounds connected to information gain in Gaussian process models.

Optimization

Kakade has studied convex optimization and non-covex optimization in machine learning. His work includes the analysis of optimization algorithms for escaping saddle points in non-convex problems. He has also co-authored research on optimization methods used in modern machine learning system.

Awards

Kakade was a co-recipient of the Test of Time Award at the International Conference on Machine Learning (ICML) in 2020 for the paper "Gaussian Process Optimization in the Bandit Setting: No Regret and Experimental Design." The ICML awards committee cited the paper's influential role in connecting Gaussian process models, bandit optimization, and experimental design.

He was a recipient of the INFORMS Revenue Management and Pricing section Prize in 2014. Kakade has served on the Alfred P. Sloan Foundation's selection committee for the Computer Science Sloan Research Fellowships.

External links

• Sham Kakade's home page
• MusicNet