Local Search is State of the Art for NAS Benchmarks

30 citations recorded by the source corpus.

How researchers see this paper

The researchers whose trusted networks rank it highest, strongest advocate first. Every researcher gets a different order.

BANANAS: Bayesian Optimization with Neural Architectures for Neural Architecture Search (2019, methodology): “[66], who showed that local search is a state-of-the-art approach on NASBench-101 and NASBench-201.”
Pitfalls in Machine Learning Research: Reexamining the Development Cycle (2020, unknown): “Despite the fact that hundreds of NAS papers have been published since Xie et al. [55], subsequent research has continued to validate the fact that basic null models based on random search remain the “state-of-the-art” for NAS [28, 52, 34].”
A Study on Encodings for Neural Architecture Search (2020, methodology): “For perturb architecture, we run experiments on regularized evolution [13] and local search [23].”
NAS-Bench-301 and the Case for Surrogate Benchmarks for Neural Architecture Search (2020, background): “For instance, local search achieves state-of-the-art results on current benchmarks but performs poorly on the search space used by DARTS (White et al., 2020b).”