Sharan Vaswani

Stochastic gradient descent (SGD) is the standard optimization method for training machine learning (ML) models. SGD requires a step-size that depends on unknown problem-dependent quantities, and the choice of this step-size heavily influences the...