Kumar Kshitij Patel
PhD student, |
I am a fourth year PhD student at Toyota Technological Institute at Chicago (TTIC) advised by Prof. Nati Srebro and Prof. Lingxiao Wang. I am interested in understanding optimization algorithms in practically relevant settings, such as settings with distributed computation and non-convexity. I am interested in understanding the min-max oracle complexity of optimization for these settings specially in presence of data and systems heterogeneity. Recently, I have been studying the game-theoretic considerations for collaboration protocols, such as cross-device federated learning. And also exploring some online robust combinatorial problems with Prof. Rad Niazadeh at the Chicago Booth School.
During summer 2020, I worked with the amazing team at Codeguru, Amazon Web Services as an applied scientist intern. And before joining TTIC, I obtained my BTech in Computer Science and Engineering at Indian Institute of Technology, Kanpur. There I was fortunate to work with Prof. Purushottam Kar on Bandit Learning algorithms. I also spent a year of my undergraduate on academic exchange at École Polytechnique Fédérale de Lausanne (EPFL) where I worked at the Machine Learning and Optimization Laboratory (MLO) with Prof. Martin Jaggi.
[Curriculum Vitae] [Google Scholar] I am co-organizing a workshop on theoritical advances in Federated Learning this summer at TTIC. I am also looking for research internships for 2024.On Convexity and Linear Mode Connectivity in Neural Networks
David Yunis, Kumar Kshitij Patel, Pedro Savarese, Karen Livescu, Matthew Walter, Jonathan Frankle, Michael Maire
OPT ML Workshop, NeurIPS 2022
Distributed Online and Bandit Convex Optimization
Kumar Kshitij Patel, Aadrirupa Saha, Lingxiao Wang, Nathan Srebro
OPT ML Workshop, NeurIPS 2022
Towards Optimal communication complexity in Distributed Non-convex Optimization [Recorded Talk]
Kumar Kshitij Patel*, Lingxiao Wang*, Blake Woodworth, Brian Bullins, Nathan Srebro (*Equal Contribution)
NeurIPS 2022
A Stochastic Newton Algorithm for Distributed Convex Optimization [Recorded Talk]
Brian Bullins, Kumar Kshitij Patel, Ohad Shamir, Nathan Srebro, Blake Woodworth (Alphabetical ordering)
NeurIPS 2021
Minibatch vs Local SGD for Heterogeneous Distributed Learning [Recorded Talk]
Blake Woodworth, Kumar Kshitij Patel, Nathan Srebro
NeurIPS 2020
Is Local SGD Better than Minibatch SGD?
Blake Woodworth, Kumar Kshitij Patel, Sebastian U. Stich, Zhen Dai, Brian Bullins, H. Brendan McMahan, Ohad Shamir, Nathan Srebro
ICML 2020
Don't Use Large Mini-batches, Use Local SGD [Code]
Tao Lin, Sebastian U. Stich, Kumar Kshitij Patel, Martin Jaggi
ICLR 2020
Communication Trade-offs for Local-SGD with Large Step Size
Kumar Kshitij Patel, Aymeric Dieuleveut
NeurIPS 2019
Corruption-Tolerant Bandit Learning
Sayash Kapoor, Kumar Kshitij Patel, Purushottam Kar
Springer Machine Learning Journal 2019
I served/am serving as a reviewer for STOC'21, TMLR, ICML'21'22, NeurIPS'21'22, ICLR'22'23, AISTATS'22'23, Springer MLJ, as a session chair for ICML'22, NeurIPS'22, and as a volunteer for ICML'20, ICLR'20. I received the top reviewer award at ICLR'22, ICML'22, NeurIPS'22.
I am participating in the NSF-Simon's research collaboration on the Mathematics of Deep Learning (MoDL).
I co-organized the TTIC Student Workshop 2021, with Gene Li. We also organized a TTIC/Uchicago student theory seminar in Spring 2021. If you'd like to take over and re-start this series, please let me know.
I was a Teaching Assistant for the Convex Optmization course at TTIC and a co-organizer for the Research at TTIC Colloquium for Fall-Winter 2021.
I participated in the Machine Learning Summer School at Tübingen, Germany during summer 2020.