I am a doctoral fellow within the Max Planck ETH Center for Learning Systems, where I am advised by Thomas Hofmann at ETH Zürich (home base) and Bernhard Schölkopf at MPI-IS Tübingen. My research interests are in the broad areas of deep learning theory, optimization, and causal representation learning.
Previously, I finished my Master’s degree in Data Science at EPFL, where I worked on model fusion and natural language processing via optimal transport, advised by Martin Jaggi. I did my master thesis at IST Austria with Dan Alistarh and focussed on efficient second-order approximation for compressing neural networks. Before EPFL, I completed my undergraduate studies in Computer Science at Indian Institute of Technology (IIT) Roorkee.
For more details on my research,
check out this research statement nay, that’s a bit too old. Please see the publications page instead. I’d be delighted to hear in case you’ve similar interests and would like to collaborate.
Pro bono: I am happy to mentor motivated Master/Bachelor students who’d like to have a taste of research in these topics (click to see some potential directions).
But, due to time constraints, please get in touch only when you are seriously interested.
Ph.D. in Computer Science (ongoing)
ETH Zürich, Max Planck Institute for Intelligent Systems
M.Sc. in Data Science, 2020
Ecole Polytechnique Federale de Lausanne
B.Tech in Computer Science, 2017
Indian Institute of Technology, Roorkee
Our first work on understanding Transformers to appear at NeurIPS’22.
September 1st, 2022: Back to Zürich 🇨🇭 — Alles Guet!
Attended the Les Houches summer school on Statistical Physics of Machine Learning — Amazing lectures + Pristine location + Cool people to hang out.
💥 @ICLR’22: Double Descent is now proved in the general case of fully-connected Neural Networks. No strict assumptions about the structure or the optimizer needed — Hessian is All You Need 😋
Nominated as a PhD student member of the European Lab for Learning and Intelligent Systems (ELLIS).
💥 Excited to share the first paper from my PhD research @ NeurIPS’21. TLDR: Neural Networks provably have much lower number of effective parameters — beautiful formulae included 🎉
Attended the Heidelberg Laureate Forum (amongst the top 225 young researchers selected in Maths and CS worldwide).
September 1st, 2021: Started my CLS exchange in the charming Tübingen 😄.
Participated in the Princeton Deep Learning Theory Summer School 2021 (really interesting although virtual).
Gave a talk on Model Fusion at the DLCT reading group (slides).
Invited talk at the Google sparsity reading group on WoodFisher (slides).
Top 33% reviewer, certificate of appreciation ICML 2020.
September 1st, 2020: Moved to Zürich for my PhD! :)
Participated in the Cornell, Maryland, Max Planck pre-doctoral research school 2020.
The preprint based on my master thesis is online, WoodFisher: Efficient second-order approximations for model compression.
Our Context Mover’s Distance paper is accepted at AISTATS 2020.
Presented our work on Model Fusion via Optimal Transport at the OTML workshop in NeurIPS (2019).
Received travel award & selected amongst top 50% reviewers, NeurIPS (2019).
Our paper on Context Mover’s Distance & Barycenters is accepted at ICLR DeepGenStruct workshop (2019).
Preprint of my internship work at FAIR is out, GLOSS: Generative Latent Optimization of Sentence Representations (2019).
Preprint for our recent paper on Wasserstein is all you need (2018).
Student grant to attend ML4P workshop in Oxford (2018).
Excited to join Facebook AI Research (FAIR) as an intern this fall (2018).
Selected for The Alan Turing Institute’s, data study group (2017).
Offered fulltime Research Fellow position at Microsoft Research India (2017).
Received Honda Y-E-S (Young Engineers and Scientists) Award 2016 (one of 14 students selected all over India).
Disclaimer: The ‘2020 Reflection(s)’ refer to only my own personal views! (also, serves as an amusement)