Inertial Proximal Deep Learning Alternating Minimization for Efficient Neutral Network Training

Abstract

In recent years, the Deep Learning Alternating Minimization (DLAM), which is actually the alternating minimization applied to the penalty form of the deep neutral networks training, has been developed as an alternative algorithm to overcome several drawbacks of Stochastic Gradient Descent (SGD) algorithms. This work develops an improved DLAM by the well-known inertial technique, namely iPDLAM, which predicts a point by linearization of current and last iterates. To obtain further training speed, we apply a warmup technique to the penalty parameter, that is, starting with a small initial one and increasing it in the iterations. Numerical results on real-world datasets are reported to demonstrate the efficiency of our proposed algorithm.

A Theoretical Framework for Inference Learning

“…However, [10] do not connect their algorithm to implicit SGD. A more recent variant of this algorithm by Qiao et al [35] show some formal links to proximal operators, an optimization process related to implicit SGD [34] (see appendix B.1). However, they do not interpret the weight updates as performing the proximal update or implicit SGD as we do here with IL.…”