Mayank Shrivastava

I am a first-year PhD student in Computer Science at the University of Illinois at Urbana-Champaign, where I am advised by Dr. Arindam Banerjee. I previously completed my master’s at UIUC, collaborating with Dr. Arindam Banerjee at UIUC, as well as with Dr. Sanmi Koyejo and Berivan Isik at Stanford University.

Before joining UIUC, I served as a research assistant under Prof. Jonathan Scarlett at the National University of Singapore, where I worked on Multi-Armed Bandits. I also spent a year at Samsung’s AI R&D Lab in South Korea, focusing on speech processing and language models.

I completed my undergraduate degree in Electrical Engineering from IIT Kanpur, where I had the opportunity to work with Dr. Himanshu Tyagi at IISc, Dr. Vipul Arora at IIT Kanpur, and Dr. Minseok Kwon at Samsung Electronics.

My current research interests lie in machine learning algorithms and theory, with a focus on federated learning algorithms, stochastic optimization for overparametrized models, and sampling algorithms.

news

Sep'24	Paper on Sketching for Distributed Learning accepted t NeuRIPS 2024!.
Aug'24	Started as a PhD student in Computer Science program at University of Illinois, Urbana-Champaign!
Mar'24	Paper on Sketching for Distributed Learning accepted in BGPT Workshop, ICLR’24.
Feb'23	Paper on Max-Quantile Bandits accepted in ALT’23.
Aug'22	Joined Master’s, Computer Science (thesis track) program at University of Illinois, Urbana-Champaign!
Apr'22	Joined Prof. Jonathan Scarlett’s lab at NUS as a Research Assistant.
Oct'20	Joined the Automatic Speech Recognition team at Samsung Electronics, South Korea as a MLE !
May'20	Started as a Research Associate at IISc Bangalore with Prof. Himanshu Tyagi.
Apr'20	Graduated with a major in Electrical Engineering from IIT Kanpur .

selected publications

2023

Max-Quantile Grouped Infinite-Arm Bandits

Ivan Lau, Yan Hao Ling, Mayank Shrivastava, and 1 more author

In Proceedings of The 34th International Conference on Algorithmic Learning Theory, 20 feb–23 feb 2023

Abs PDF

In this paper, we consider a bandit problem in which there are a number of groups each consisting of infinitely many arms. Whenever a new arm is requested from a given group, its mean reward is drawn from an unknown reservoir distribution (different for each group), and the uncertainty in the arm’s mean reward can only be reduced via subsequent pulls of the arm. The goal is to identify the infinite-arm group whose reservoir distribution has the highest (1-α)-quantile (e.g., median if α= \frac12), using as few total arm pulls as possible. We introduce a two-step algorithm that first requests a fixed number of arms from each group and then runs a finite-arm grouped max-quantile bandit algorithm. We characterize both the instance-dependent and worst-case regret, and provide a matching lower bound for the latter, while discussing various strengths, weaknesses, algorithmic improvements, and potential lower bounds associated with our instance-dependent upper bounds.