Machine learning researcher and engineer specializing in large-scale sequential modeling, optimization geometry, low-precision training numerics, and statistical inference. https://ekansharma.org/ Fri, 13 Nov 2026 23:14:49 -0500 Fri, 13 Nov 2026 23:14:49 -0500 Our preprint TokenMinds: Pretrained User Tokens and Embeddings for User Understanding in Large Recommender Systems is now available on arXiv. https://ekansharma.org/news/#june-2026 https://ekansharma.org/news/#june-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. https://ekansharma.org/news/#may-2026 https://ekansharma.org/news/#may-2026 Ph.D. in Computer Science conferred at the University of Toronto. Thesis: Exploiting linear connectivity modulo permutations in neural networks for parsimonious machine learning. https://ekansharma.org/news/#november-2025 https://ekansharma.org/news/#november-2025 Started at Google as a Machine Learning Engineer working on large-scale recommender systems for YouTube Shorts. https://ekansharma.org/news/#september-2025 https://ekansharma.org/news/#september-2025 My joint work with Mohammed Adnan, Rohan Jain, Rahul G. Krishnan, and Yani Ioannou on Sparse Training from Random Initialization: Aligning Lottery Ticket Masks using Weight Symmetry is accepted to ICML 2025. https://ekansharma.org/news/#may-2025 https://ekansharma.org/news/#may-2025 Entered the job market following my Ph.D. research (archived milestone — see my updated CV). https://ekansharma.org/news/#january-2025 https://ekansharma.org/news/#january-2025 My joint work with Mohammed Adnan, Rohan Jain, and Yani Ioannou on Winning Tickets from Random Initialization: Aligning Masks for Sparse Training is accepted to the UniREPS workshop at NeurIPS 2024. https://ekansharma.org/news/#december-2024 https://ekansharma.org/news/#december-2024 New preprint on The Non-Local Model Merging Problem: Permutation Symmetries and Variance Collapse is out on arXiv, with code open-sourced in tact-merge. https://ekansharma.org/news/#october-2024 https://ekansharma.org/news/#october-2024 My joint work with Devin Kwok, Tom Denton, Daniel M. Roy, David Rolnick, and Gintare K. Dziugaite on Simultaneous Linear Connectivity of Neural Networks Modulo Permutation is accepted to ECML PKDD 2024. https://ekansharma.org/news/#may-2024 https://ekansharma.org/news/#may-2024 Our work on Simultaneous Linear Connectivity of Neural Networks Modulo Permutation was accepted for a non-archival spotlight at CPAL 2024. https://ekansharma.org/news/#november-2023 https://ekansharma.org/news/#november-2023