Neil Rohit Mallinar
UC San Diego. Halıcıoğlu Data Science Institute. nmallina@ucsd.edu.

3rd floor
HDSI
La Jolla, CA 92093
I’m pursuing my PhD at UC San Diego, advised by Misha Belkin. I am also a PhD Research Fellow at The Eric and Wendy Schmidt Center of The Broad Institute where I was adivsed by Anthony Philippakis (now a General Partner at GV). For a full list of publications, see Google Scholar.
My research focuses on understanding deep learning, both theoretically and practically. In the past, I studied: benign overfitting in neural networks and kernel regression; calibration in neural networks as it relates to generalization; spectral properties of the de Bruijn graph Laplacian; high-dimensional covariate shifts; and emergent phenomena in non-neural models through grokking modular arithmetic.
I am currently researching topics on fundamentals of feature learning, foundation model training and evaluation (through the lens of single-cell data), scaling laws and feature alignment, and multi-class learning in kernels and neural networks.
news
Jun 10, 2024 | Joined Google Research (Mountain View) as a PhD Research Intern |
---|---|
Jun 05, 2023 | Joined Microsoft Research New England (MSR-NE) as a PhD Research Intern |
Jul 06, 2022 | Attended the Summer Cluster on Deep Learning Theory at Simons Institute for Theory of Computing from July - August, 2022. |
Jan 01, 2022 | Honored to be supported in my PhD through funding from the Eric and Wendy Schmidt Center at The Broad Institute of MIT & Harvard. |
Sep 28, 2020 | Started my PhD at UC San Diego |
Aug 01, 2019 | Joined Pryon Inc in Brooklyn as an AI Research Engineer |
May 18, 2016 | Got my B.S. from Johns Hopkins University in Computer Science and Mathematics |
selected publications
-
- Emergence in non-neural models: grokking modular arithmetic via average gradient outer productarXin preprint, Oct 2024
- Eigenvectors of the De Bruijn Graph Laplacian: A Natural Basis for the Cut and Cycle SpacearXiv preprint, Oct 2024
- Minimum-Norm Interpolation Under Covariate ShiftIn Proceedings of the 41st International Conference on Machine Learning (ICML 2024), Mar 2024
- Benign, tempered, or catastrophic: A taxonomy of overfittingIn 36th Conference on Neural Information Processing Systems (NeurIPS 2022), Jul 2022