How Does Supernet Help in Neural Architecture Search?
This work addresses the debate over weight sharing's benefits in NAS, providing insights for researchers to optimize its use, though it is incremental as it builds on existing methods.
The paper investigates the effectiveness of weight sharing in Neural Architecture Search (NAS) by analyzing its performance across five search spaces, finding it works well in some but fails in others, and identifies biases to explain these results.
Weight sharing, as an approach to speed up architecture performance estimation has received wide attention. Instead of training each architecture separately, weight sharing builds a supernet that assembles all the architectures as its submodels. However, there has been debate over whether the NAS process actually benefits from weight sharing, due to the gap between supernet optimization and the objective of NAS. To further understand the effect of weight sharing on NAS, we conduct a comprehensive analysis on five search spaces, including NAS-Bench-101, NAS-Bench-201, DARTS-CIFAR10, DARTS-PTB, and ProxylessNAS. We find that weight sharing works well on some search spaces but fails on others. Taking a step forward, we further identified biases accounting for such phenomenon and the capacity of weight sharing. Our work is expected to inspire future NAS researchers to better leverage the power of weight sharing.