Local Critic Training of Deep Neural Networks

TL;DR

This paper introduces a new training method for deep neural networks using local critic networks to enable layer-wise training without full backpropagation, improving efficiency and flexibility.

Contribution

It presents a novel local critic training approach with cascaded learning, enhancing neural network training, inference, and ensemble performance.

Findings

Effective training without full backpropagation

Improved computational efficiency and inference

Enhanced ensemble classification performance

Abstract

This paper proposes a novel approach to train deep neural networks by unlocking the layer-wise dependency of backpropagation training. The approach employs additional modules called local critic networks besides the main network model to be trained, which are used to obtain error gradients without complete feedforward and backward propagation processes. We propose a cascaded learning strategy for these local networks. In addition, the approach is also useful from multi-model perspectives, including structural optimization of neural networks, computationally efficient progressive inference, and ensemble classification for performance improvement. Experimental results show the effectiveness of the proposed approach and suggest guidelines for determining appropriate algorithm parameters.