MutualNet: Adaptive ConvNet via Mutual Learning from Network Width and Resolution
This work addresses the need for efficient and adaptive neural networks in resource-constrained environments, offering a novel training strategy that is not incremental but provides broad performance gains across multiple tasks.
The paper tackles the problem of training neural networks to adapt dynamically to resource constraints by proposing MutualNet, a method that trains sub-networks with varying widths and resolutions to learn multi-scale representations. It achieves better ImageNet accuracy than state-of-the-art adaptive networks, outperforming US-Net and EfficientNet by up to 1.5%, and shows improvements in object detection, segmentation, and transfer learning.
We propose the width-resolution mutual learning method (MutualNet) to train a network that is executable at dynamic resource constraints to achieve adaptive accuracy-efficiency trade-offs at runtime. Our method trains a cohort of sub-networks with different widths using different input resolutions to mutually learn multi-scale representations for each sub-network. It achieves consistently better ImageNet top-1 accuracy over the state-of-the-art adaptive network US-Net under different computation constraints, and outperforms the best compound scaled MobileNet in EfficientNet by 1.5%. The superiority of our method is also validated on COCO object detection and instance segmentation as well as transfer learning. Surprisingly, the training strategy of MutualNet can also boost the performance of a single network, which substantially outperforms the powerful AutoAugmentation in both efficiency (GPU search hours: 15000 vs. 0) and accuracy (ImageNet: 77.6% vs. 78.6%). Code is available at \url{https://github.com/taoyang1122/MutualNet}.