NASTransfer: Analyzing Architecture Transferability in Large Scale Neural Architecture Search

Rameswar Panda, Michele Merler, Mayoore Jaiswal, Hui Wu, Kandan Ramakrishnan, Ulrich Finkler, Chun-Fu Chen, Minsik Cho, Rogerio Feris, David Kung, Bishwaranjan Bhattacharjee

Abstract: Neural Architecture Search (NAS) is an open and challeng- ing problem in machine learning. While NAS offers great promise, the prohibitive computational demand of most of the existing NAS methods makes it difficult to directly search the architectures on large-scale tasks. The typical way of con- ducting large scale NAS is to search for an architectural build- ing block on a small dataset (either using a proxy set from the large dataset or a completely different small scale dataset) and then transfer the block to a larger dataset. Despite a number of recent results that show the promise of transfer from proxy datasets, a comprehensive evaluation of different NAS meth- ods studying the impact of different source datasets has not yet been addressed. In this work, we propose to analyze the architecture transferability of different NAS methods by per- forming a series of experiments on large scale benchmarks such as ImageNet1K and ImageNet22K. We find that: (i) The size and domain of the proxy set does not seem to influ- ence architecture performance on the target dataset. On av- erage, transfer performance of architectures searched using completely different small datasets (e.g., CIFAR10) perform similarly to the architectures searched directly on proxy tar- get datasets. However, design of proxy sets has considerable impact on rankings of different NAS methods. (ii) While dif- ferent NAS methods show similar performance on a source dataset (e.g., CIFAR10), they significantly differ on the trans- fer performance to a large dataset (e.g., ImageNet1K). (iii) Even on large datasets, random sampling baseline is very competitive, but the choice of the appropriate combination of proxy set and search strategy can provide significant improve- ment over it. We believe that our extensive empirical analysis will prove useful for future design of NAS algorithms.