Training-free Neural Architecture Search through Variance of Knowledge of Deep Network Weights

TL;DR

This paper introduces a training-free proxy based on Fisher Information for neural architecture search, significantly reducing computational costs while achieving state-of-the-art accuracy estimates without training networks.

Contribution

It proposes a novel Fisher Information-based proxy for NAS that eliminates the need for training, backed by strong theoretical foundations and validated on multiple datasets.

Findings

State-of-the-art results on three datasets

Effective in two different search spaces

Proposed a new, more informative metric for NAS

Abstract

Deep learning has revolutionized computer vision, but it achieved its tremendous success using deep network architectures which are mostly hand-crafted and therefore likely suboptimal. Neural Architecture Search (NAS) aims to bridge this gap by following a well-defined optimization paradigm which systematically looks for the best architecture, given objective criterion such as maximal classification accuracy. The main limitation of NAS is however its astronomical computational cost, as it typically requires training each candidate network architecture from scratch. In this paper, we aim to alleviate this limitation by proposing a novel training-free proxy for image classification accuracy based on Fisher Information. The proposed proxy has a strong theoretical background in statistics and it allows estimating expected image classification accuracy of a given deep network without training the network, thus significantly reducing computational cost of standard NAS algorithms. Our training-free proxy achieves state-of-the-art results on three public datasets and in two search spaces, both when evaluated using previously proposed metrics, as well as using a new metric that we propose which we demonstrate is more informative for practical NAS applications. The source code is publicly available at http://www.github.com/ondratybl/VKDNW