System and method for providing a hardware-specific machine learning model architecture

A system, comprising:

• one or more processors;
• and
• one or more non-transitory storage devices storing instructions for providing a hardware-specific machine learning model architecture, the instructions, when executed by the one or more processors, causing the one or more processors to perform the following steps including:
• obtaining one or more architectural parameters of a machine learning model architecture, wherein the machine learning model architecture includes one or more model architecture elements, and the one or more architectural parameters include one or more architectural parameters of model architecture elements in the one or more model architecture elements,
• simulating an execution of the machine learning model architecture on a given target hardware, and obtaining from the simulating one or more metrics describing a performance, by defining a hardware efficiency, on the given target hardware, of individual model architecture elements out of the one or more model architecture elements,
• for each respective architectural parameter of the one or more architectural parameters:
• identifying one or more model architecture elements of the machine learning model architecture impacted by adjusting the respective architectural parameter,
• determining a mutation score for the respective architectural parameter from the one or more metrics for the one or more identified model architecture elements, 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 machine learning model architecture on the given target hardware,
• using the mutation scores for the one or more architectural parameters in a genetic neural architecture search algorithm, wherein an input of the genetic neural architecture search algorithm includes the machine learning model architecture, and the mutation scores indicate a probability that the respective architectural parameters are adjusted, and wherein an output of the genetic neural architecture search algorithm includes an adjusted machine learning model architecture, and
• providing the adjusted machine learning model architecture as output.

A non-transitory computer-readable medium on which are stored data representing instructions for providing a hardware-specific machine learning model architecture, the instructions, when executed by a processor system, causing the processor system to perform the following steps including:

• obtaining one or more architectural parameters of a machine learning model architecture, wherein the machine learning model architecture includes one or more model architecture elements, and the one or more architectural parameters include one or more architectural parameters of model architecture elements in the one or more model architecture elements,
• simulating an execution of the machine learning model architecture on a given target hardware, and obtaining from the simulating one or more metrics describing a performance, by defining a hardware efficiency, on the given target hardware, of individual model architecture elements out of the one or more model architecture elements,
• for each respective architectural parameter of the one or more architectural parameters:
• identifying one or more model architecture elements of the machine learning model architecture impacted by adjusting the respective architectural parameter,
• determining a mutation score for the respective architectural parameter from the one or more metrics for the one or more identified model architecture elements, 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 machine learning model architecture on the given target hardware,
• using the mutation scores for the one or more architectural parameters in a genetic neural architecture search algorithm, wherein an input of the genetic neural architecture search algorithm includes the machine learning model architecture, and the mutation scores indicate a probability that the respective architectural parameters are adjusted, and wherein an output of the genetic neural architecture search algorithm includes an adjusted machine learning model architecture, and
• providing the adjusted machine learning model architecture as output.