Improved Techniques for Training Adaptive Deep Networks

TL;DR

This paper introduces new training techniques for adaptive deep networks, enhancing their efficiency and collaboration among classifiers, leading to better performance on standard datasets.

Contribution

It proposes three novel training methods to improve the efficacy of adaptive deep networks, focusing on classifier collaboration and conflict resolution.

Findings

Improved training efficacy on CIFAR-10, CIFAR-100, and ImageNet.

Enhanced efficiency of adaptive networks with proposed techniques.

Consistent performance gains over state-of-the-art methods.

Abstract

Adaptive inference is a promising technique to improve the computational efficiency of deep models at test time. In contrast to static models which use the same computation graph for all instances, adaptive networks can dynamically adjust their structure conditioned on each input. While existing research on adaptive inference mainly focuses on designing more advanced architectures, this paper investigates how to train such networks more effectively. Specifically, we consider a typical adaptive deep network with multiple intermediate classifiers. We present three techniques to improve its training efficacy from two aspects: 1) a Gradient Equilibrium algorithm to resolve the conflict of learning of different classifiers; 2) an Inline Subnetwork Collaboration approach and a One-for-all Knowledge Distillation algorithm to enhance the collaboration among classifiers. On multiple datasets (CIFAR-10, CIFAR-100 and ImageNet), we show that the proposed approach consistently leads to further improved efficiency on top of state-of-the-art adaptive deep networks.