Machine Learning Engineer · Google (YouTube Shorts)

Ekansh Sharma

Ph.D. in Computer Science, University of Toronto

I am a Machine Learning Engineer at Google working on large-scale recommender systems for YouTube Shorts. I completed my Ph.D. in the Department of Computer Science at the University of Toronto, where I was advised by Daniel M. Roy and affiliated with the Vector Institute. I also hold an M.Sc. in Computer Science and a B.A.Sc. with Honors in Electrical Engineering from the University of Toronto.

My work spans machine learning methodology, optimization geometry, and statistical inference. At Google, I build LLM-adapted generative sequential user models, long-context compression bottlenecks, low-precision (FP8/INT4) quantization-aware training pipelines, and scaling-law methodology in JAX/XLA. During my Ph.D., I investigated the geometry of neural network loss landscapes—studying linear mode connectivity modulo permutation symmetries, non-local model merging, and sparse optimization dynamics—to design parsimonious, compute-conscious learning algorithms.

Earlier in my research, I studied probabilistic programming, Bayesian nonparametrics, and heavy-tailed random graph estimation, and spent a term as an Applied Scientist Intern on Sponsored Display Ads at Amazon.

Google Scholar GitHub CV (PDF)
Portrait of Ekansh Sharma
Mountain View, CA
Focus Large-Scale Sequential & Recommender Modeling Loss-Landscape Geometry & Model Merging Sparse & Low-Precision Optimization Statistical & Causal Inference

Publications & Preprints

Preprint 2026 Sequential User Modeling Recommender Systems

TokenMinds: Pretrained User Tokens and Embeddings for User Understanding in Large Recommender Systems

Qingyun Liu, Bo Yan, Yang Liu, Yuji Roh, Ekansh Sharma, Likang Yin, Emma Olowo, Yuxuan Li, Yuan Jiang, Diego Uribe, Saksham Aggarwal, Siqi Wu, Yuan Hao, Vikas Kedigehalli, Min-Hsuan Tsai, Lukasz Heldt, Lichan Hong, Li Wei, Xinyang Yi

Preprint arXiv:2606.25147 · 2026

arXiv PDF
ICML 2026 Sparse Training Optimization

SparseOpt: Addressing Normalization-induced Gradient Skew in Sparse Training

Mohammed Adnan, Rohan Jain, Tom Jacobs, Ekansh Sharma, Rahul G. Krishnan, Rebekka Burkholz, Yani Ioannou

International Conference on Machine Learning (ICML) · 2026

arXiv PDF
ICML 2025 Sparse Training Weight Symmetry

Sparse Training from Random Initialization: Aligning Lottery Ticket Masks using Weight Symmetry

Mohammed Adnan, Rohan Jain, Ekansh Sharma, Rahul G. Krishnan, Yani Ioannou

International Conference on Machine Learning (ICML) · 2025

Preprint 2024 Model Merging Loss Landscape

The Non-Local Model Merging Problem: Permutation Symmetries and Variance Collapse

Ekansh Sharma, Daniel M. Roy, Gintare Karolina Dziugaite

Preprint arXiv:2410.12766 · 2024

arXiv PDF Code
ECML-PKDD 2024 CPAL 2024 Spotlight Mode Connectivity Permutation Symmetries

Simultaneous Linear Connectivity of Neural Networks Modulo Permutation

Ekansh Sharma, Devin Kwok, Tom Denton, Daniel M. Roy, David Rolnick, Gintare Karolina Dziugaite

ECML PKDD 2024 · CPAL 2024 Spotlight Track

Paper arXiv PDF DOI
EJS 2021 Random Graphs Heavy-Tailed Asymptotics

Bootstrap estimators for the tail-index and for the count statistics of graphex processes

Zacharie Naulet, Daniel M. Roy, Ekansh Sharma, Victor Veitch

Electronic Journal of Statistics, 15(1), 282–325 · 2021

Journal arXiv PDF DOI
PROBPROG 2020 Probabilistic Programming Nonasymptotic MCMC

Approximations in Probabilistic Programs: A Compositional Nonasymptotic Analysis of Nested MCMC

Ekansh Sharma, Daniel M. Roy

International Conference on Probabilistic Programming (PROBPROG) · 2020

arXiv PDF Talk Poster
Preprint 2020 Relational Data Bayesian Nonparametrics

Exchangeable modelling of relational data: checking sparsity, train-test splitting, and sparse exchangeable Poisson matrix factorization

Victor Veitch, Ekansh Sharma, Zacharie Naulet, Daniel M. Roy

Preprint arXiv:1712.02311 · 2020

arXiv PDF

Recent News

All news →
May 2026
My joint work with Mohammed Adnan, Rohan Jain, Tom Jacobs, Rahul G. Krishnan, Rebekka Burkholz, and Yani Ioannou on SparseOpt: Addressing Normalization-induced Gradient Skew in Sparse Training is accepted to ICML 2026.
November 2025
Ph.D. in Computer Science conferred at the University of Toronto. Thesis: Exploiting linear connectivity modulo permutations in neural networks for parsimonious machine learning.
September 2025
Started at Google as a Machine Learning Engineer working on large-scale recommender systems for YouTube Shorts.