BlockDrop: Dynamic Inference Paths in Residual Networks

TL;DR

BlockDrop is a method that dynamically selects which residual blocks to evaluate in a deep neural network during inference, significantly reducing computation while maintaining high accuracy, by learning policies through reinforcement learning.

Contribution

It introduces a reinforcement learning-based framework to adaptively drop residual blocks in ResNets during inference, achieving faster predictions without accuracy loss.

Findings

Achieves up to 36% speedup on ImageNet with no accuracy degradation.

Learned policies encode meaningful visual information.

Method is effective on CIFAR and ImageNet datasets.

Abstract

Very deep convolutional neural networks offer excellent recognition results, yet their computational expense limits their impact for many real-world applications. We introduce BlockDrop, an approach that learns to dynamically choose which layers of a deep network to execute during inference so as to best reduce total computation without degrading prediction accuracy. Exploiting the robustness of Residual Networks (ResNets) to layer dropping, our framework selects on-the-fly which residual blocks to evaluate for a given novel image. In particular, given a pretrained ResNet, we train a policy network in an associative reinforcement learning setting for the dual reward of utilizing a minimal number of blocks while preserving recognition accuracy. We conduct extensive experiments on CIFAR and ImageNet. The results provide strong quantitative and qualitative evidence that these learned policies not only accelerate inference but also encode meaningful visual information. Built upon a ResNet-101 model, our method achieves a speedup of 20\% on average, going as high as 36\% for some images, while maintaining the same 76.4\% top-1 accuracy on ImageNet.