Praneeth Netrapalli
 |
I am a senior researcher at Microsoft Research India, Bengaluru. I am also an adjunct professor at TIFR, Mumbai and a faculty associate of ICTS, Bengaluru. Prior to this, I was a postdoctoral researcher at Microsoft Research New England in Cambridge, MA. I obtained my MS and PhD in ECE from UT Austin where I was fortunate to be advised by Sujay Sanghavi. Before coming to UT, I worked in the Quantitative Analysis group at Goldman Sachs, Bangalore. I obtained my B-Tech in Electrical Engineering from IIT Bombay. Honors and awards:
|
Research Interests
The main focus of my work is to develop a deeper understanding of machine learning, ranging across optimization, generalization, robustness etc. Given the difficulty and diversity of these questions, I use both theoretical [Paper1 Paper2] and experimental [Paper3] tools to make progress on these questions.
I also work on designing efficient algorithms for various optimization problems that arise in machine learning [Paper4].
News and announcements
Our paper showing the optimality of gradient descent for nonsmooth convex optimization, even in the quantum setting has been accepted to ITCS 2021. Joint work with Ankit Garg, Robin Kothari and Suhail Sherif.
Five papers accepted to NeurIPS 2020. One on pitfalls of simplicity bias in neural networks, one on least squares regression with Markovian data, one on model based offline RL, one on incorporating optimism in FTPL and finally one on projection efficient algorithms for nonsmooth and stochastic convex optimization.
Two papers accepted to ICML 2020. One on nonconvex-nonconcave minimax optimization and another on domain generalization.
Arun Suggala won the best student paper award at ALT 2020 for our paper Online Non-Convex Learning: Following the Perturbed Leader is Optimal. Congrats Arun!
Our paper Phase Retrieval using Alternating Minimization has been selected for the IEEE Signal Processing Society Best Paper Award for 2019. This award is given annually to up to six papers published during the previous five years. Thanks to my coauthors Sujay Sanghavi and Prateek Jain for making this work possible.
Professional service
On the program committee/area chair: NeurIPS 2019-20, COLT 2019-21, ICLR 2021, ICML 2019, 2021, AISTATS 2019.
Selected publications
* indicates alphabetical ordering of author names. Please see here for all publications.
How to Escape Saddle Points Efficiently
C. Jin, R. Ge, P. Netrapalli, S. M. Kakade and M. I. Jordan
ICML 2017Streaming PCA: Matching Matrix Bernstein and Near-Optimal Finite Sample Guarantees for Oja's Algorithm [*]
P. Jain, C. Jin, S. M. Kakade, P. Netrapalli and A. Sidford
COLT 2016Phase Retrieval using Alternating Minimization
P. Netrapalli, P. Jain and S. Sanghavi
IEEE Transactions on Signal Processing 2015, Vol. 63, Issue 18, Pages 4814–4826Low-rank Matrix Completion using Alternating Minimization [*]
P. Jain, P. Netrapalli and S. Sanghavi
STOC 2013
Mentoring
I have been extremely fortunate to work with an amazing set of students.
Interns/Visitors (in reverse chronological order)
Suhas Kowshik (MIT)
Carrie Wu (Stanford)
Vihari Piratla (IIT Bombay)
Kaustav Tamuly (BITS Goa)
Suhail M. Shah
Sheshansh Agrawal (IIT Bombay; now at MSR India)
Arun Sai Suggala (CMU)
Dheeraj Nagaraj (MIT)
Govinda M. Kamath (Stanford; now postdoc at MSR New England)
Rahul Kidambi (Univ. of Washington; now at Amazon)
Chi Jin (UC Berkeley; now Asst. Prof. at Princeton EE)
Research fellows (in reverse chronological order)
Abhishek Panigrahi
Suhas Jayaram (now PhD student at CMU)
Raghav Somani (now PhD student at University of Washington)
Chirag Gupta (now PhD student at CMU)
Rahul Anand Sharma (now PhD student at CMU)
Vivek Gupta (now PhD student at University of Utah)
Yeshwanth Cherapanamjeri (now PhD student at UC Berkeley)
Contact details
Address: Microsoft Research, # 9, Vigyan 1st floor, Lavelle Road, Bengaluru, Karnataka 560001, India.
E-mail: praneeth [at] microsoft [dot] com