papers
Publications (2)
cs.DC2020
Holdout SGD: Byzantine Tolerant Federated Learning
Shahar Azulay, Lior Raz, Amir Globerson +2
This work presents a new distributed Byzantine tolerant federated learning algorithm, HoldOut SGD, for Stochastic Gradient Descent (SGD) optimization. HoldOut SGD uses the well kno…
cs.LG2021
On the Implicit Bias of Initialization Shape: Beyond Infinitesimal Mirror Descent
Shahar Azulay, Edward Moroshko, Mor Shpigel Nacson +4
Recent work has highlighted the role of initialization scale in determining the structure of the solutions that gradient methods converge to. In particular, it was shown that large…