Syamantak Kumar

I am a third year PhD student studying computer science at UT Austin advised by Prof. Purnamrita Sarkar and Prof. Kevin Tian . Broadly, my research interests lie at the intersection of statistics and machine learning. I am interested in designing machine learning algorithms with provable guarantees of convergence and correctness. I am also interested to work on quantifying the computational and statistical aspects of state-of-the-art deep learning methods. I deeply enjoy studying the applications of high-dimensional statistics, optimization and probability theory to real-world problems. Previously, I was an undergraduate student at the Computer Science and Engineering Department, IIT Bombay .

I previously worked with Prof. Suyash Awate on Medical Imaging and with Prof. Preethi Jyothi on NLP for Code-switching at IIT Bombay. I have also worked with Prof. Thomas Deserno on Medical Informatics and Imaging, back in 2018.

Email  /  CV  /  Github  /  Google Scholar  /  LinkedIn

• [September 2025]   Selected for the Amazon AI PhD Fellowship!
• [Summer 2024]   Worked with Dheeraj on Score Matching for Diffusion models as a research intern at Google Deepmind, Bengaluru
• [Sept 2023]   Our paper "Streaming PCA for Markovian Data" accepted as a Spotlight (Top ~3%) at NeurIPS 2023.
• [Aug 2023]   Started PhD CS at UT Austin.
• [Aug 2021]   Started MS CS at UT Austin.
• [Jan 2021]   Our paper "From Machine Translation to Code-Switching : Generating High-Quality Code-Switched Text" accepted in ACL-IJCNLP 2021.
• [July 2020]   Joined Google, Bangalore as a software engineer in the Google Maps team.
• [Jun 2019]   Working at Google, Bangalore as a software engineering intern.
• [Feb 2019]   Our paper "A comparison of open source libraries ready for 3D reconstruction of wounds" got accepted in SPIE Medical Imaging 2019.

We study Oja's algorithm for streaming PCA under linear and nonlinear stochastic quantization and show that a batched version achieves the lower bound up to logarithmic factors under both schemes.

We establish the first (nearly) dimension-free sample complexity bounds for learning these score functions, achieving a double exponential improvement in dimension over prior results. Building on these insights, we propose Bootstrapped Score Matching (BSM), a variance reduction technique that utilizes previously learned scores to improve accuracy at higher noise levels.

We give an near-linear time differentially-private algorithm for computing the geometric median.

We give the first provable algorithms for spike-and-slab posterior sampling that apply for any SNR, and use a measurement count sublinear in the problem dimension.

We propose a novel statistical inference framework for streaming principal component analysis (PCA) using Oja's algorithm, enabling the construction of confidence intervals for individual entries of the estimated eigenvector.

We analyze Oja's algorithm with thresholding for streaming sparse PCA, achieving near-optimal rates of convergence.

We develop a nonparametric score to adaptively pick the right algorithm for Independent Component Analysis (ICA) with arbitrary Gaussian noise

We theoretically analyze deflation-based algorithms for k-PCA using 1-PCA oracles.

We propose a nearly-optimal streaming algorithm for performing PCA under Markovian dependence among the samples.

In this work, we adapt a state-of-the-art neural machine translation model to generate Hindi-English code-switched sentences starting from monolingual Hindi sentences.

We propose an android application for real-time three dimensional scanning of surface wounds to aid effective diagnosis of wounds remotely and present a comparative study of open-source libraries available for performing this task.

I worked at Google, Bangalore with the Maps team on a model for predicting the correctness of an edit made by a user on Maps. This model is used to ensure that data displayed on Google Maps is true to the accuracy it claims.

Used Generative Adversarial Networks (GANs) to generate adversarial examples for keyword spotting systems, improving their robustness.

Designed an algorithm for fully-sequential sampling in a Probably-Approximately-Correct (PAC) setting to determine top-K players in a tournament, given pariwise preferences

Implemented an AI for playing Chinese Checkers using the Minimax algorithm with alpha-beta pruning. (Implemented in Racket)