Tanmay Gautam

Tanmay Gautam 

Ph.D. Candidate in the Department of Electrical Engineering and Computer Sciences
University of California, Berkeley

Advisor: Somayeh Sojoudi

Contact

Email: tgautam23 (at) berkeley.edu

LinkedIn / Resume

Research Interests

About Me

I am a third-year Ph.D. student in EECS at UC Berkeley, where I am fortunate to be advised by Professor Somayeh Sojoudi. I'm broadly interested in working at the intersection of machine learning and optimization. More specifically, I'm interested in applying concepts from optimization theory to tackle various challenges in the domains of reinforcement learning and deep learning. Within reinforcement learning, some topics that excite me include safe RL, inverse RL and also working on establishing theoretical guarantees (e.g. convergence, sample efficiency) for current RL algorithms. Within the realm of deep learning, I'm particularly fascinated by the notion of implicit layers (see here for exposition) and various instantiations thereof including Deep Equilibrium Models (DEQs), Neural Ordinary Differential Equations (ODEs) and differentiable optimization. For an overview of my past projects see the publication section below.

Previously, I received my undergraduate degree from Imperial College London, where I completed the 4-year integrated MEng. degree in Electrical and Electronic Engineering. While at Imperial, I had the opportunity to work on a research project within the Control and Power (CAP) Research Group in Imperial College London, under supervision of Professor Alessandro Astolfi and Dr. Giordano Scarciotti. This was an independent project on model reduction - a field wherein various mathematical techniques have been developed to produce a simplified mathematical description of an already existing large-order dynamical system. Furthermore, I carried out my MEng. thesis project under supervision of Professor Patrick Naylor in the Speech and Audio Processing (SAP) Laboratory at Imperial College London. This project was focused on the research domain of speaker diarization.

Publications

Preprints

  1. A Sequential Greedy Approach for Training Implicit Deep Models, 2022, under review
         Tanmay Gautam, Brendon G. Anderson, Somayeh Sojoudi and Laurent El Ghaoui

  2. Efficient Global Optimization of Two-layer ReLU Networks: Quadratic-time Algorithms and Adversarial Training, 2022, under review
         Yatong Bai, Tanmay Gautam and Somayeh Sojoudi

Conference Papers

  1. Safe Reinforcement Learning with Chance-constrained Model Predictive Control, Learning for Dynamics and Control Conference, 2022
         Samuel Pfrommer, Tanmay Gautam, Alec Zhou and Somayeh Sojoudi

  2. Practical Convex Formulation of Robust One-hidden-layer Neural Network Training, American Control Conference, 2022
         Yatong Bai, Tanmay Gautam, Yu Gai and Somayeh Sojoudi

Education

Coursework



UC Berkeley 

UC Berkeley by night. (Source: Visit California)