System und Verfahren zur Bereitstellung einer Hardwarespezifischen Maschinenlernmodellarchitektur

Details

Veröffentlichungs-Nr.

EP4787241

Anmeldetag

31. Januar 2025

Veröffentlichung

5. August 2026

Rechtsraum

EP

Abstract

A method for providing a hardware-architecture-specific machine learning model architecture 300 is provided. One or more architectural parameters 321-341 of a machine learning model architecture are obtained, the model architecture comprising model architecture elements 310-360. An execution of the model architecture is simulated on a given target hardware, from which one or more metrics 322-342 describing a performance of the elements are obtained. For each architectural parameter, model architecture elements are identified which are impacted by adjusting the respective architectural parameter. A mutation score 323-343 is determined for the architectural parameter from the one or more metrics, the mutation score indicating an impact of the one or more model architecture elements associated with the respective architectural parameter on an overall performance of the model architecture on the given target hardware. The mutation scores are used in a genetic neural architecture search algorithm, indicating probabilities that architectural parameters are adjusted.

Anmelder

Firma

Robert Bosch GmbH

Land

🇩🇪 Deutschland