Width Transfer: On the (In)variance of Width Optimization

TL;DR

This paper introduces width transfer, a method to drastically reduce the computational overhead of width optimization in CNNs by leveraging the invariance of optimal widths across different sizes and depths, maintaining accuracy.

Contribution

The paper proposes width transfer, a novel technique that minimizes width optimization overhead and demonstrates the invariance of optimal widths across architectures and datasets.

Findings

Achieves up to 320x reduction in width optimization overhead.

Maintains top-1 accuracy on ImageNet despite reduced overhead.

Shows widths for better accuracy are invariant across architectures and training data.

Abstract

Optimizing the channel counts for different layers of a CNN has shown great promise in improving the efficiency of CNNs at test-time. However, these methods often introduce large computational overhead (e.g., an additional 2x FLOPs of standard training). Minimizing this overhead could therefore significantly speed up training. In this work, we propose width transfer, a technique that harnesses the assumptions that the optimized widths (or channel counts) are regular across sizes and depths. We show that width transfer works well across various width optimization algorithms and networks. Specifically, we can achieve up to 320x reduction in width optimization overhead without compromising the top-1 accuracy on ImageNet, making the additional cost of width optimization negligible relative to initial training. Our findings not only suggest an efficient way to conduct width optimization but also highlight that the widths that lead to better accuracy are invariant to various aspects of network architectures and training data.