Surprisingly Strong Performance Prediction with Neural Graph Features

TL;DR

This paper introduces neural graph features (GRAF), a simple and interpretable method for fast performance prediction in neural architecture search that outperforms existing zero-cost proxies and predictors.

Contribution

The paper proposes GRAF, a novel graph-based feature set for performance prediction that is efficient, interpretable, and more accurate than existing zero-cost proxies.

Findings

GRAF outperforms zero-cost proxies in performance prediction.

Combining GRAF with other proxies yields superior results.

GRAF is computationally efficient and interpretable.

Abstract

Performance prediction has been a key part of the neural architecture search (NAS) process, allowing to speed up NAS algorithms by avoiding resource-consuming network training. Although many performance predictors correlate well with ground truth performance, they require training data in the form of trained networks. Recently, zero-cost proxies have been proposed as an efficient method to estimate network performance without any training. However, they are still poorly understood, exhibit biases with network properties, and their performance is limited. Inspired by the drawbacks of zero-cost proxies, we propose neural graph features (GRAF), simple to compute properties of architectural graphs. GRAF offers fast and interpretable performance prediction while outperforming zero-cost proxies and other common encodings. In combination with other zero-cost proxies, GRAF outperforms most existing performance predictors at a fraction of the cost.