Noisy Differentiable Architecture Search

7 May 2020  ·  Xiangxiang Chu, Bo Zhang ·

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.

PDF Abstract
Task Dataset Model Metric Name Metric Value Global Rank Result Benchmark
Neural Architecture Search CIFAR-10 NoisyDARTS Top-1 Error Rate 2.39% # 12
Search Time (GPU days) 0.25 # 8
Parameters 3.25M # 22
FLOPS 534M # 37
Image Classification CIFAR-10 NoisyDARTS-A-t Percentage correct 98.28 # 43
Image Classification CIFAR-10 NoisyDARTS-a Percentage correct 97.61 # 69
PARAMS 5.5M # 196
Neural Architecture Search ImageNet NoisyDARTS-A Top-1 Error Rate 22.1 # 54
Accuracy 77.9 # 43
Params 5.5M # 30
MACs 449M # 115

Methods


No methods listed for this paper. Add relevant methods here