NAS-Bench-1Shot1: Benchmarking and Dissecting One-shot Neural Architecture Search
This work provides a tool for researchers to systematically study and improve one-shot NAS methods, addressing a bottleneck in making NAS computationally feasible, though it is incremental as it builds on existing benchmarks.
The paper tackles the lack of understanding in one-shot neural architecture search (NAS) by introducing a general benchmarking framework, NAS-Bench-1Shot1, based on NAS-Bench-101, enabling cheap evaluations and analysis of methods, and uses it to compare state-of-the-art methods, showing sensitivity to hyperparameters and performance relative to blackbox optimizers.
One-shot neural architecture search (NAS) has played a crucial role in making NAS methods computationally feasible in practice. Nevertheless, there is still a lack of understanding on how these weight-sharing algorithms exactly work due to the many factors controlling the dynamics of the process. In order to allow a scientific study of these components, we introduce a general framework for one-shot NAS that can be instantiated to many recently-introduced variants and introduce a general benchmarking framework that draws on the recent large-scale tabular benchmark NAS-Bench-101 for cheap anytime evaluations of one-shot NAS methods. To showcase the framework, we compare several state-of-the-art one-shot NAS methods, examine how sensitive they are to their hyperparameters and how they can be improved by tuning their hyperparameters, and compare their performance to that of blackbox optimizers for NAS-Bench-101.