ExpandNets: Linear Over-parameterization to Train Compact Convolutional Networks
arXiv:1811.10495
Abstract
We introduce an approach to training a given compact network. To this end, we leverage over-parameterization, which typically improves both neural network optimization and generalization. Specifically, we propose to expand each linear layer of the compact network into multiple consecutive linear layers, without adding any nonlinearity. As such, the resulting expanded network, or ExpandNet, can be contracted back to the compact one algebraically at inference. In particular, we introduce two convolutional expansion strategies and demonstrate their benefits on several tasks, including image classification, object detection, and semantic segmentation. As evidenced by our experiments, our approach outperforms both training the compact network from scratch and performing knowledge distillation from a teacher. Furthermore, our linear over-parameterization empirically reduces gradient confusion during training and improves the network generalization.
References in corpus (13)
- Distilling the Knowledge in a Neural Network
- MobileNets: Efficient Convolutional Neural Networks for Mobile Vision Applications
- YOLOv3: An Incremental Improvement
- FitNets: Hints for Thin Deep Nets
- Paying More Attention to Attention: Improving the Performance of Convolutional Neural Networks via Attention Transfer
- Simplifying Graph Convolutional Networks
- A Convergence Theory for Deep Learning via Over-Parameterization
- SNIP: Single-shot Network Pruning based on Connection Sensitivity
- YOLO9000: Better, Faster, Stronger
- Speeding-up Convolutional Neural Networks Using Fine-tuned CP-Decomposition
- Generalization in Deep Learning
- Compression-aware Training of Deep Networks
- The Impact of Neural Network Overparameterization on Gradient Confusion and Stochastic Gradient Descent