BlockDrop: Dynamic Inference Paths in Residual Networks
arXiv:1711.08393
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.
arXiv commentCVPR 2018
References in corpus (26)
- Very Deep Convolutional Networks for Large-Scale Image Recognition
- Distilling the Knowledge in a Neural Network
- MobileNets: Efficient Convolutional Neural Networks for Mobile Vision Applications
- SqueezeNet: AlexNet-level accuracy with 50x fewer parameters and <0.5MB model size
- Deep Residual Learning for Image Recognition
- Deep Compression: Compressing Deep Neural Networks with Pruning, Trained Quantization and Huffman Coding
- R-FCN: Object Detection via Region-based Fully Convolutional Networks
- Binarized Neural Networks: Training Deep Neural Networks with Weights and Activations Constrained to +1 or -1
- FitNets: Hints for Thin Deep Nets
- Residual Networks Behave Like Ensembles of Relatively Shallow Networks
- Compressing Neural Networks with the Hashing Trick
- Aggregated Residual Transformations for Deep Neural Networks
- Convolutional neural networks with low-rank regularization
- Adaptive Computation Time for Recurrent Neural Networks
- Deep Networks with Stochastic Depth
- Conditional Computation in Neural Networks for faster models
- Deep Learning of Representations: Looking Forward
- BranchyNet: Fast Inference via Early Exiting from Deep Neural Networks
- An exploration of parameter redundancy in deep networks with circulant projections
- Multi-Residual Networks: Improving the Speed and Accuracy of Residual Networks
- More is Less: A More Complicated Network with Less Inference Complexity
- Deep Sequential Neural Network
- Quantized Convolutional Neural Networks for Mobile Devices
- Data-Driven Sparse Structure Selection for Deep Neural Networks
- Spatially Adaptive Computation Time for Residual Networks
- SEP-Nets: Small and Effective Pattern Networks
Cited by in corpus (6)
- Slimmable Neural Networks
- Pixel-wise Attentional Gating for Parsimonious Pixel Labeling
- Improved Techniques for Training Adaptive Deep Networks
- Anytime Inference with Distilled Hierarchical Neural Ensembles
- TAFE-Net: Task-Aware Feature Embeddings for Low Shot Learning
- Anytime Stereo Image Depth Estimation on Mobile Devices