Deep Neural Networks with Short Circuits for Improved Gradient Learning

TL;DR

This paper introduces short circuit neural connections that enhance gradient flow in deep neural networks, reducing reliance on external training and improving performance across vision and language tasks.

Contribution

It proposes a novel unidirectional short circuit connection that improves gradient learning without adding external parameters.

Findings

Significant performance gains over baselines in vision tasks

Effective gradient enhancement in NLP applications

No additional external training parameters needed

Abstract

Deep neural networks have achieved great success both in computer vision and natural language processing tasks. However, mostly state-of-art methods highly rely on external training or computing to improve the performance. To alleviate the external reliance, we proposed a gradient enhancement approach, conducted by the short circuit neural connections, to improve the gradient learning of deep neural networks. The proposed short circuit is a unidirectional connection that single back propagates the sensitive from the deep layer to the shallows. Moreover, the short circuit formulates to be a gradient truncation of its crossing layers which can plug into the backbone deep neural networks without introducing external training parameters. Extensive experiments demonstrate deep neural networks with our short circuit gain a large margin over the baselines on both computer vision and natural language processing tasks.