Residual Networks of Residual Networks: Multilevel Residual Networks

TL;DR

This paper introduces Residual networks of Residual networks (RoR), a multilevel residual architecture that enhances the optimization and performance of deep residual networks across various datasets.

Contribution

The paper proposes a novel multilevel residual architecture, RoR, which improves residual network optimization and can be applied to different residual network variants.

Findings

RoR significantly boosts residual network performance.

RoR achieves state-of-the-art results on CIFAR-10, CIFAR-100, and SVHN.

RoR outperforms ResNets on ImageNet dataset.

Abstract

A residual-networks family with hundreds or even thousands of layers dominates major image recognition tasks, but building a network by simply stacking residual blocks inevitably limits its optimization ability. This paper proposes a novel residual-network architecture, Residual networks of Residual networks (RoR), to dig the optimization ability of residual networks. RoR substitutes optimizing residual mapping of residual mapping for optimizing original residual mapping. In particular, RoR adds level-wise shortcut connections upon original residual networks to promote the learning capability of residual networks. More importantly, RoR can be applied to various kinds of residual networks (ResNets, Pre-ResNets and WRN) and significantly boost their performance. Our experiments demonstrate the effectiveness and versatility of RoR, where it achieves the best performance in all residual-network-like structures. Our RoR-3-WRN58-4+SD models achieve new state-of-the-art results on CIFAR-10, CIFAR-100 and SVHN, with test errors 3.77%, 19.73% and 1.59%, respectively. RoR-3 models also achieve state-of-the-art results compared to ResNets on ImageNet data set.