BlockDrop: Dynamic Inference Paths in Residual Networks
This addresses the problem of high computational costs for real-world applications of deep networks, offering a practical speedup without accuracy loss, though it is incremental as it builds on existing ResNet architectures.
The paper tackles the computational expense of deep convolutional neural networks by introducing BlockDrop, a method that dynamically selects which layers to execute during inference, achieving an average speedup of 20% (up to 36%) while maintaining 76.4% top-1 accuracy on ImageNet.
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.