Tandem Blocks in Deep Convolutional Neural Networks
This work addresses a fundamental understanding gap in neural network design for researchers and practitioners, but it is incremental as it builds on existing residual network architectures.
The paper investigates the role of shortcut connections in deep convolutional neural networks, hypothesizing that they work as linear counterparts to nonlinear layers, and finds that alternative linear connections can outperform identity shortcuts, with effectiveness varying by network width and depth.
Due to the success of residual networks (resnets) and related architectures, shortcut connections have quickly become standard tools for building convolutional neural networks. The explanations in the literature for the apparent effectiveness of shortcuts are varied and often contradictory. We hypothesize that shortcuts work primarily because they act as linear counterparts to nonlinear layers. We test this hypothesis by using several variations on the standard residual block, with different types of linear connections, to build small image classification networks. Our experiments show that other kinds of linear connections can be even more effective than the identity shortcuts. Our results also suggest that the best type of linear connection for a given application may depend on both network width and depth.