Angelos Assos

abstract:Deep neural networks have shown great success in many machine learning tasks. Their training is challenging since the loss surface of the network architecture is generally non-convex, or even non-smooth. Thus, there has been a long-standing question on how optimization algorithms may converge to a global minimum. In general, existing optimization algorithms cannot guarantee the convergence to a global solution without convexity or other strong assumptions. In this project, we study optimization frameworks for machine learning tasks and design (stochastic) gradient-based methods to solve this problem. We aim to provide theoretical guarantees for these methods with convergence to global solutions using reasonable assumptions.