In this paper, we perform Lyapunov based analysis of the loss function to derive an a priori upper bound on the settling time of deep neural networks. While previous studies have attempted to understand deep learning using control theory framework, there is limited work on a priori finite time convergence analysis. Drawing from the advances in analysis of finite-time control of non-linear systems, we provide a priori guarantees of finite-time convergence in a deterministic control theoretic setting. We formulate the supervised learning framework as a control problem where weights of the network are control inputs and learning translates into a tracking problem. An analytical formula for finite-time upper bound on settling time is computed a priori under the assumptions of boundedness of input. Finally, we prove the robustness and sensitivity of the loss function against input perturbations.
@article{arxiv.2009.07509,
title = {A priori guarantees of finite-time convergence for Deep Neural Networks},
author = {Anushree Rankawat and Mansi Rankawat and Harshal B. Oza},
journal= {arXiv preprint arXiv:2009.07509},
year = {2020}
}