Scalable parameter encoding of artificial neural networks obtained via an evolutionary process

US10599975B2 · kind B2 · utility

Assignee

• Uber Technologies, Inc. · US

Inventors

• Felipe Petroski Such · San Francisco, US
• Jeffrey Michael Clune · San Francisco, US
• Kenneth O. Stanley · San Francisco, US
• Edoardo Conti · San Francisco, US
• Vashisht Madhavan · San Francisco, US
• Joel Anthony Lehman · San Francisco, US

Key dates

• Filing date: Dec 14, 2018
• Grant date: Mar 24, 2020
• Expiry date: Dec 14, 2038

Classification

• Technology area (CPC G): Physics
• CPC primary: G06N3/126
• WIPO field: Computer technology
• WIPO sector: Electrical engineering

Abstract

A source system initializes, using an initialization seed, a first parameter vector representing weights of a neural network. The source system determines a second parameter vector by performing a sequence of mutations on the first parameter vector, the mutations each being based on a perturbation seed. The source system generates, and stores to memory, an encoded representation of the second parameter vector that comprises the initialization seed and a sequence of perturbation seeds corresponding to the sequence of mutations. The source system transmits the data structure to a target system, which processes a neural network based on the data structure.