Noisy Differentiable Architecture Search

TL;DR

NoisyDARTS introduces random noise into differentiable architecture search to prevent performance collapse caused by skip connections, smoothing the loss landscape and achieving state-of-the-art results across multiple tasks.

Contribution

It proposes a simple, effective noise injection method to improve DARTS, addressing the skip connection issue without biasing the gradient, and provides theoretical and empirical validation.

Findings

Achieves state-of-the-art results on various benchmarks

Demonstrates noise smoothing of the loss landscape

Robustly improves architecture search stability

Abstract

Simplicity is the ultimate sophistication. Differentiable Architecture Search (DARTS) has now become one of the mainstream paradigms of neural architecture search. However, it largely suffers from the well-known performance collapse issue due to the aggregation of skip connections. It is thought to have overly benefited from the residual structure which accelerates the information flow. To weaken this impact, we propose to inject unbiased random noise to impede the flow. We name this novel approach NoisyDARTS. In effect, a network optimizer should perceive this difficulty at each training step and refrain from overshooting, especially on skip connections. In the long run, since we add no bias to the gradient in terms of expectation, it is still likely to converge to the right solution area. We also prove that the injected noise plays a role in smoothing the loss landscape, which makes the optimization easier. Our method features extreme simplicity and acts as a new strong baseline. We perform extensive experiments across various search spaces, datasets, and tasks, where we robustly achieve state-of-the-art results. Our code is available at https://github.com/xiaomi-automl/NoisyDARTS.