ISTA-NAS: Efficient and Consistent Neural Architecture Search by Sparse Coding

TL;DR

This paper introduces ISTA-NAS, a neural architecture search method that formulates the search as a sparse coding problem, enabling efficient, consistent, and one-stage search and evaluation, achieving state-of-the-art results with minimal computational cost.

Contribution

The paper proposes a novel sparse coding formulation for NAS, allowing efficient differentiable search in a compressed space and unified architecture search and evaluation in one stage.

Findings

Two-stage method requires only 0.05 GPU-day on CIFAR-10.

One-stage method achieves state-of-the-art results on CIFAR-10 and ImageNet.

Efficient training and consistent architecture evaluation are demonstrated.

Abstract

Neural architecture search (NAS) aims to produce the optimal sparse solution from a high-dimensional space spanned by all candidate connections. Current gradient-based NAS methods commonly ignore the constraint of sparsity in the search phase, but project the optimized solution onto a sparse one by post-processing. As a result, the dense super-net for search is inefficient to train and has a gap with the projected architecture for evaluation. In this paper, we formulate neural architecture search as a sparse coding problem. We perform the differentiable search on a compressed lower-dimensional space that has the same validation loss as the original sparse solution space, and recover an architecture by solving the sparse coding problem. The differentiable search and architecture recovery are optimized in an alternate manner. By doing so, our network for search at each update satisfies the sparsity constraint and is efficient to train. In order to also eliminate the depth and width gap between the network in search and the target-net in evaluation, we further propose a method to search and evaluate in one stage under the target-net settings. When training finishes, architecture variables are absorbed into network weights. Thus we get the searched architecture and optimized parameters in a single run. In experiments, our two-stage method on CIFAR-10 requires only 0.05 GPU-day for search. Our one-stage method produces state-of-the-art performances on both CIFAR-10 and ImageNet at the cost of only evaluation time.