ExpandNets: Linear Over-parameterization to Train Compact Convolutional Networks
This method addresses the challenge of efficiently training compact networks for deployment in resource-constrained environments, though it is incremental as it builds on existing over-parameterization techniques.
The authors tackled the problem of training compact convolutional networks by introducing ExpandNets, which linearly over-parameterize layers to improve optimization and generalization, resulting in outperformance over training from scratch and knowledge distillation on tasks like image classification, object detection, and semantic segmentation.
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.