Device placement optimization with reinforcement learning

US10692003B2 · kind B2 · utility

Assignee

• Google LLC · US

Inventors

• Samuel Bengio · Los Altos, US
• Mohammad Norouzi · Toronto, CA
• Benoit Steiner · Mountain View, US
• Jeffrey Adgate Dean · Palo Alto, US
• Hieu Hy Pham · Redwood City, US
• Azalia Mirhoseini · San Jose, US
• Quoc V. Le · Stanford, US
• Naveen Kumar · San Mateo, US
• Yuefeng Zhou · Haywards Heath, GB
• Rasmus Larsen · San Jose, US

Key dates

• Filing date: Jun 19, 2019
• Grant date: Jun 23, 2020
• Expiry date: Jun 19, 2039

Classification

• Technology area (CPC G): Physics
• CPC primary: G06N5/04
• WIPO field: Computer technology
• WIPO sector: Electrical engineering

Abstract

A method for determining a placement for machine learning model operations across multiple hardware devices is described. The method includes receiving data specifying a machine learning model to be placed for distributed processing on multiple hardware devices; generating, from the data, a sequence of operation embeddings, each operation embedding in the sequence characterizing respective operations necessary to perform the processing of the machine learning model; processing the sequence of operation embeddings using a placement recurrent neural network in accordance with first values of a plurality network parameters of the placement recurrent neural network to generate a network output that defines a placement of the operations characterized by the operation embeddings in the sequence across the plurality of devices; and scheduling the machine learning model for processing by the multiple hardware devices by placing the operations on the multiple devices according to the placement defined by the network output.