BaLeNAS: Differentiable Architecture Search via the Bayesian Learning Rule

TL;DR

This paper introduces BaLeNAS, a Bayesian learning-based differentiable architecture search method that models architecture parameters as distributions, improving stability and exploration, and achieves state-of-the-art results on standard benchmarks.

Contribution

It formulates neural architecture search as a distribution learning problem using NGVI, enhancing exploration and stability in differentiable NAS.

Findings

Significant performance improvements on NAS-Bench-201 and NAS-Bench-1shot1.

Achieves state-of-the-art results using training-free proxies for architecture selection.

Obtains competitive test errors on CIFAR-10, CIFAR-100, and ImageNet datasets.

Abstract

Differentiable Architecture Search (DARTS) has received massive attention in recent years, mainly because it significantly reduces the computational cost through weight sharing and continuous relaxation. However, more recent works find that existing differentiable NAS techniques struggle to outperform naive baselines, yielding deteriorative architectures as the search proceeds. Rather than directly optimizing the architecture parameters, this paper formulates the neural architecture search as a distribution learning problem through relaxing the architecture weights into Gaussian distributions. By leveraging the natural-gradient variational inference (NGVI), the architecture distribution can be easily optimized based on existing codebases without incurring more memory and computational consumption. We demonstrate how the differentiable NAS benefits from Bayesian principles, enhancing exploration and improving stability. The experimental results on NAS-Bench-201 and NAS-Bench-1shot1 benchmark datasets confirm the significant improvements the proposed framework can make. In addition, instead of simply applying the argmax on the learned parameters, we further leverage the recently-proposed training-free proxies in NAS to select the optimal architecture from a group architectures drawn from the optimized distribution, where we achieve state-of-the-art results on the NAS-Bench-201 and NAS-Bench-1shot1 benchmarks. Our best architecture in the DARTS search space also obtains competitive test errors with 2.37\%, 15.72\%, and 24.2\% on CIFAR-10, CIFAR-100, and ImageNet datasets, respectively.