SMASH: One-Shot Model Architecture Search through HyperNetworks

TL;DR

SMASH introduces a hypernetwork-based approach to rapidly search for effective neural network architectures by generating weights conditioned on architecture, enabling efficient exploration of diverse designs with minimal training runs.

Contribution

The paper presents a novel HyperNet technique that accelerates neural architecture search by generating weights conditioned on architecture, allowing broad exploration with a single training.

Findings

Achieved competitive performance on CIFAR-10, CIFAR-100, STL-10, ModelNet10, and Imagenet32x32.

Enabled efficient architecture search with a single training run.

Flexible mechanism for defining various network connectivity patterns.

Abstract

Designing architectures for deep neural networks requires expert knowledge and substantial computation time. We propose a technique to accelerate architecture selection by learning an auxiliary HyperNet that generates the weights of a main model conditioned on that model's architecture. By comparing the relative validation performance of networks with HyperNet-generated weights, we can effectively search over a wide range of architectures at the cost of a single training run. To facilitate this search, we develop a flexible mechanism based on memory read-writes that allows us to define a wide range of network connectivity patterns, with ResNet, DenseNet, and FractalNet blocks as special cases. We validate our method (SMASH) on CIFAR-10 and CIFAR-100, STL-10, ModelNet10, and Imagenet32x32, achieving competitive performance with similarly-sized hand-designed networks. Our code is available at https://github.com/ajbrock/SMASH