SiGeo: Sub-One-Shot NAS via Information Theory and Geometry of Loss Landscape

TL;DR

SiGeo introduces a novel sub-one-shot NAS framework that leverages limited training data and information theory to efficiently evaluate neural architectures, outperforming existing proxies with reduced computational costs.

Contribution

The paper proposes a new sub-one-shot NAS paradigm and a theoretical proxy, SiGeo, that effectively bridges zero-shot and one-shot NAS, reducing training requirements and computational costs.

Findings

SiGeo outperforms state-of-the-art NAS proxies on benchmarks.

Warm-up with SiGeo achieves comparable performance to traditional one-shot NAS.

Sub-one-shot NAS reduces computational costs by approximately 60%.

Abstract

Neural Architecture Search (NAS) has become a widely used tool for automating neural network design. While one-shot NAS methods have successfully reduced computational requirements, they often require extensive training. On the other hand, zero-shot NAS utilizes training-free proxies to evaluate a candidate architecture's test performance but has two limitations: (1) inability to use the information gained as a network improves with training and (2) unreliable performance, particularly in complex domains like RecSys, due to the multi-modal data inputs and complex architecture configurations. To synthesize the benefits of both methods, we introduce a "sub-one-shot" paradigm that serves as a bridge between zero-shot and one-shot NAS. In sub-one-shot NAS, the supernet is trained using only a small subset of the training data, a phase we refer to as "warm-up." Within this framework, we present SiGeo, a proxy founded on a novel theoretical framework that connects the supernet warm-up with the efficacy of the proxy. Extensive experiments have shown that SiGeo, with the benefit of warm-up, consistently outperforms state-of-the-art NAS proxies on various established NAS benchmarks. When a supernet is warmed up, it can achieve comparable performance to weight-sharing one-shot NAS methods, but with a significant reduction ($\sim 60$\%) in computational costs.