SAMARTH SINHA

samarth [dot] sinha [at] mail [dot] utoronto [dot] ca

I am an undergraduate student in Computer Engineering at the University of Toronto and Vector Institute where I work with Animesh Garg and Roger Grosse on the intersection of machine learning and reinforcement learning.

Previously, I was a visiting student at MILA working closely with Anirudh Goyal, Hugo Larochelle and Yoshua Bengio; I was also a visitor at UC Berkeley, working with Trevor Darrell and Zeynep Akata. I have also been very fortunate to work at some amazing startups: Aurora, DeepMap, Concured, Nanoleaf.

I spent my high school summers working with Alex Lvovsky working on Quantum Optics and with Guenther Ruhe working on Software Engineering at the University of Calgary.

Twitter

Google Scholar

RESEARCH INTERESTS

My research interests span across Deep Learning and (Deep) Reinforcement Learning. I am interested in training models and agents that can learn generalizable representations from minimal supervision.

I have had the privilege to work with some amazing collaborators, and I am always interested in chatting about new problems. If you find anything interesting on the website, please get in touch!

Get in Touch

PUBLICATIONS / PRE-PRINTS

DIVA: DIVERSE VISUAL FEATURE AGGREGATION FOR DEEP METRIC LEARNING

Karsten Roth*, Timo Milbich*, Homanga Bharadhwaj, Samarth Sinha, Joseph Paul Cohen, Bjorn Ommer, Yoshua Bengio (2020)

ECCV 2020

CURRICULUM BY SMOOTHING

Samarth Sinha, Animesh Garg, Hugo Larochelle (2020)

arXiv

TOP-K TRAINING OF GANS: IMPROVING GAN PERFORMANCE BY THROWING AWAY BAD SAMPLES

Samarth Sinha*, Zhengli Zhao*, Anirudh Goyal, Colin Raffel, Augustus Odena (2020)

arXiv

EXPERIENCE REPLAY WITH LIKELIHOOD-FREE IMPORTANCE WEIGHTS

Samarth Sinha*, Jiaming Song*, Animesh Garg, Stefano Ermon (2020)

arXiv

REVISITING TRAINING STRATEGIES AND GENERALIZATION PERFORMANCE IN DEEP METRIC LEARNING

Karsten Roth*, Timo Milbich*, Samarth Sinha, Prateek Gupta, Bjoern Ommer, Joseph Paul Cohen (2020)

ICML 2020

DIBS: DIVERSITY INDUCING INFORMATION BOTTLENECK IN MODEL ENSEMBLES

Samarth Sinha*, Homanga Bharadhwaj*, Anirudh Goyal, Hugo Larochelle, Animesh Garg, Florian Shkurti (2020)

ICML-UDL Workshop 2020, arXiv

SAMARTH SINHA

RESEARCH INTERESTS

PUBLICATIONS / PRE-PRINTS

DIVA: DIVERSE VISUAL FEATURE AGGREGATION FOR DEEP METRIC LEARNING

CURRICULUM BY SMOOTHING

TOP-K TRAINING OF GANS: IMPROVING GAN PERFORMANCE BY THROWING AWAY BAD SAMPLES

EXPERIENCE REPLAY WITH LIKELIHOOD-FREE IMPORTANCE WEIGHTS

REVISITING TRAINING STRATEGIES AND GENERALIZATION PERFORMANCE IN DEEP METRIC LEARNING

DIBS: DIVERSITY INDUCING INFORMATION BOTTLENECK IN MODEL ENSEMBLES

SMALL-GAN: SPEEDING UP GAN TRAINING USING CORE-SETS

VARIATIONAL ADVERSARIAL ACTIVE LEARNING

GENERALIZED ZERO- AND FEW-SHOT LEARNING VIA ALIGNED VARIATIONAL AUTOENCODERS

MAX ENTROPY MODELS FOR FAST ADAPTATION