sharpDARTS: Faster and More Accurate Differentiable Architecture Search

Abstract

Neural Architecture Search (NAS) has been a source of dramatic improvements in neural network design, with recent results meeting or exceeding the performance of hand-tuned architectures. However, our understanding of how to represent the search space for neural net architectures and how to search that space efficiently are both still in their infancy. We have performed an in-depth analysis to identify limitations in a widely used search space and a recent architecture search method, Differentiable Architecture Search (DARTS). These findings led us to introduce novel network blocks with a more general, balanced, and consistent design; a better-optimized Cosine Power Annealing learning rate schedule; and other improvements. Our resulting sharpDARTS search is 50% faster with a 20-30% relative improvement in final model error on CIFAR-10 when compared to DARTS. Our best single model run has 1.93% (1.98+/-0.07) validation error on CIFAR-10 and 5.5% error (5.8+/-0.3) on the recently released CIFAR-10.1 test set. To our knowledge, both are state of the art for models of similar size. This model also generalizes competitively to ImageNet at 25.1% top-1 (7.8% top-5) error. We found improvements for existing search spaces but does DARTS generalize to new domains? We propose Differentiable Hyperparameter Grid Search and the HyperCuboid search space, which are representations designed to leverage DARTS for more general parameter optimization. Here we find that DARTS fails to generalize when compared against a human's one shot choice of models. We look back to the DARTS and sharpDARTS search spaces to understand why, and an ablation study reveals an unusual generalization gap. We finally propose Max-W regularization to solve this problem, which proves significantly better than the handmade design. Code will be made available.

arXiv comment9 pages, 6 figures, 4 tables

References in corpus (9)

• Improved Regularization of Convolutional Neural Networks with Cutout
• MobileNetV2: Inverted Residuals and Linear Bottlenecks
• DARTS: Differentiable Architecture Search
• AutoAugment: Learning Augmentation Policies from Data
• Large-Scale Evolution of Image Classifiers
• SMASH: One-Shot Model Architecture Search through HyperNetworks
• Shake-Shake regularization
• Do CIFAR-10 Classifiers Generalize to CIFAR-10?
• Network Decoupling: From Regular to Depthwise Separable Convolutions

Cited by in corpus (19)

• AutoML: A Survey of the State-of-the-Art
• Differentiable Quantum Architecture Search
• NAS evaluation is frustratingly hard
• ASAP: Architecture Search, Anneal and Prune
• Weight-Sharing Neural Architecture Search: A Battle to Shrink the Optimization Gap
• XNAS: Neural Architecture Search with Expert Advice
• DASS: Differentiable Architecture Search for Sparse neural networks
• Single-Path NAS: Device-Aware Efficient ConvNet Design
• Neural Architecture Search for Deep Face Recognition
• MiLeNAS: Efficient Neural Architecture Search via Mixed-Level Reformulation
• Prune and Replace NAS
• Geometry-Aware Gradient Algorithms for Neural Architecture Search
• COVID Detection in Chest CTs: Improving the Baseline on COV19-CT-DB
• Multi-Objective Neural Architecture Search Based on Diverse Structures and Adaptive Recommendation
• Single-Path Mobile AutoML: Efficient ConvNet Design and NAS Hyperparameter Optimization
• Single-level Optimization For Differential Architecture Search
• Reconsidering CO2 emissions from Computer Vision
• Neural Architecture Search in Embedding Space
• Single-DARTS: Towards Stable Architecture Search