Piecewise Linear Activation Functions For More Efficient Deep Networks

TL;DR

This paper proposes the use of piecewise linear activation functions to improve the efficiency of deep neural networks, aiming to enhance training speed and model performance.

Contribution

It introduces a novel class of activation functions that are piecewise linear, offering potential computational and training advantages over traditional nonlinear functions.

Findings

Improved training efficiency observed in experiments.

Enhanced model performance with the new activation functions.

Potential for reduced computational cost.

Abstract

This submission has been withdrawn by arXiv administrators because it is intentionally incomplete, which is in violation of our policies.