Deep Learning Scaling is Predictable, Empirically

TL;DR

This paper empirically analyzes how generalization error and model size scale with data and computation in deep learning, revealing power-law relationships across multiple domains that inform future research and system design.

Contribution

It introduces a methodology for measuring scaling laws and provides large-scale empirical evidence of power-law error and model size scaling in deep learning.

Findings

Generalization error follows a power-law scaling across domains.

Model size scales sublinearly with data size.

Scaling relationships are consistent across different machine learning tasks.

Abstract

Deep learning (DL) creates impactful advances following a virtuous recipe: model architecture search, creating large training data sets, and scaling computation. It is widely believed that growing training sets and models should improve accuracy and result in better products. As DL application domains grow, we would like a deeper understanding of the relationships between training set size, computational scale, and model accuracy improvements to advance the state-of-the-art. This paper presents a large scale empirical characterization of generalization error and model size growth as training sets grow. We introduce a methodology for this measurement and test four machine learning domains: machine translation, language modeling, image processing, and speech recognition. Our empirical results show power-law generalization error scaling across a breadth of factors, resulting in power-law exponents---the "steepness" of the learning curve---yet to be explained by theoretical work. Further, model improvements only shift the error but do not appear to affect the power-law exponent. We also show that model size scales sublinearly with data size. These scaling relationships have significant implications on deep learning research, practice, and systems. They can assist model debugging, setting accuracy targets, and decisions about data set growth. They can also guide computing system design and underscore the importance of continued computational scaling.