The Uncanny Similarity of Recurrence and Depth

TL;DR

Recurrent neural networks, despite reusing the same parameters, exhibit hierarchical feature extraction and depth-related performance benefits similar to deep feed-forward networks, often with fewer parameters.

Contribution

This work demonstrates that recurrent models can emulate deep hierarchical behaviors and performance gains of feed-forward networks, challenging the notion that depth is solely tied to distinct layer parameters.

Findings

Recurrent networks exhibit hierarchical feature extraction similar to deep models.

Recurrent models achieve comparable performance with fewer parameters.

Recurrent networks can emulate the behavior of deep feed-forward networks.

Abstract

It is widely believed that deep neural networks contain layer specialization, wherein neural networks extract hierarchical features representing edges and patterns in shallow layers and complete objects in deeper layers. Unlike common feed-forward models that have distinct filters at each layer, recurrent networks reuse the same parameters at various depths. In this work, we observe that recurrent models exhibit the same hierarchical behaviors and the same performance benefits with depth as feed-forward networks despite reusing the same filters at every recurrence. By training models of various feed-forward and recurrent architectures on several datasets for image classification as well as maze solving, we show that recurrent networks have the ability to closely emulate the behavior of non-recurrent deep models, often doing so with far fewer parameters.