Tanmay Gangwani

I am an Applied Scientist at Amazon, working towards solving various user-centric problems via the application of machine learning concepts. I enjoy designing solutions that leverage academic and industrial research in the areas of reinforcement learning, deep learning, and natural language processing.

I received my Ph.D. in Computer Science from the University of Illinois, Urbana Champaign, where I was advised by Prof. Jian Peng. My Ph.D. thesis focused on applied algorithms for deep reinforcement learning (RL). Concretely, I studied approaches that utilize expert demonstrations for RL (imitation learning), address the issues of exploration and sparse environmental rewards, and improve sample efficiency with the transfer-RL and meta-RL paradigms.

Before this, my research was at the intersection of computer architecture and compilers. I earned my Master's degree from the same school, advised by Prof. Josep Torrellas.

Email  /  CV  /  Google Scholar  /  Twitter  /  Github  

Status: Defended my Ph.D. thesis in August 2021; joined the Amazon Science group in October 2021
Publications
Hindsight Foresight Relabeling for Meta-Reinforcement Learning
Michael Wan, Jian Peng, Tanmay Gangwani
International Conference on Learning Representations (ICLR), 2022
Imitation Learning from Observations under Transition Model Disparity
Tanmay Gangwani, Yuan Zhou, Jian Peng
International Conference on Learning Representations (ICLR), 2022
Learning Guidance Rewards with Trajectory-space Smoothing
Tanmay Gangwani, Yuan Zhou, Jian Peng
Conference on Neural Information Processing Systems (NeurIPS), 2020
code / poster / bibtex
Harnessing Distribution Ratio Estimators for Learning Agents with Quality and Diversity
Tanmay Gangwani, Jian Peng, Yuan Zhou
Conference on Robot Learning (CoRL), 2020
code / slides / video / bibtex
State-only Imitation with Transition Dynamics Mismatch
Tanmay Gangwani, Jian Peng
International Conference on Learning Representations (ICLR), 2020
code / slides / video / bibtex
Mutual Information Based Knowledge Transfer Under State-Action Dimension Mismatch
Michael Wan, Tanmay Gangwani, Jian Peng
The Conference on Uncertainty in Artificial Intelligence (UAI), 2020
slides / video / bibtex
Learning Belief Representations for Imitation Learning in POMDPs
Tanmay Gangwani, Joel Lehman, Qiang Liu, Jian Peng
The Conference on Uncertainty in Artificial Intelligence (UAI), 2019
code / bibtex
Learning Self-Imitating Diverse Policies
Tanmay Gangwani, Qiang Liu, Jian Peng
International Conference on Learning Representations (ICLR), 2019
code / slides / poster / bibtex
Policy Optimization by Genetic Distillation
Tanmay Gangwani, Jian Peng
International Conference on Learning Representations (ICLR), 2018
poster / bibtex
Distributed and Secure ML using Self-tallying Multi-party Aggregation
Yunhui Long*, Tanmay Gangwani*, Haris Mughees, Carl Gunter
(* denotes co-first authorship)
NeurIPS workshop on Privacy Preserving Machine Learning (PPML), 2018
code / slides / bibtex
Architectural Support for Relaxed Concurrent Priority Queueing in Chip Multiprocessors
Azin Heidarshenas*, Tanmay Gangwani*, Serif Yesil, Adam Morrison, Josep Torrellas
(* denotes co-first authorship)
International Conference on Supercomputing (ICS), 2020
slides / bibtex
Breaking Serialization in Lock-Free Multicore Synchronization
Tanmay Gangwani, Adam Morrison, Josep Torrellas
Architectural Support for Programming Languages and Operating Systems (ASPLOS), 2016
poster / slides / bibtex
Partial Redundancy Elimination using Lazy Code Motion
Sandeep Dasgupta, Tanmay Gangwani
Technical Report, 2015
code / bibtex

Template imitation-learning (pun intended) using this!