Compute Trends Across Three Eras of Machine Learning

Compute, data, and algorithmic advances are the three fundamental factors that guide the progress of modern Machine Learning (ML). In this Jaime Sevilla, Lennart Heim, Anson Ho, Tamay Besiroglu, Marius Hobbhahn, and Pablo Villalobos article at, 2022 International Joint Conference on Neural Networks (IJCNN), study trends in the most readily quantified factor - compute. 

They show that before 2010 training compute grew in line with Moore's law, doubling roughly every 20 months. Since the advent of Deep Learning in the early 2010s, the scaling of training compute has accelerated, doubling approximately every 6 months. In late 2015, a new trend emerged as firms developed large-scale ML models with 10 to 100-fold larger requirements in training compute. 

Based on these observations we split the history of compute in ML into three eras: the Pre Deep Learning Era, the Deep Learning Era, and the Large-Scale Era. Overall, our work highlights the fast-growing compute requirements for training advanced ML systems.

Get access to the publication