Searching compression profiles for trained neural networks

A system, comprising:

• one or more processors;
• and
• a memory, that stores program instructions that, when executed by the at least one processor, cause the one or more processors to:
• automatically search for a compression profile that optimizes a performance characteristic of one or more neural networks trained according to a respective one or more data sets, wherein the searching comprises:
• generate different prospective compression profiles for the one or more trained neural networks;
• collect performance information of the different prospective compression profiles for respectively compressed versions the one or more trained neural networks using the one or more data sets;
• evaluate the performance information of the different prospective compression profiles for the respective compressed versions of the one or more trained neural networks to identify one of the different prospective compression profiles that optimizes the performance characteristic of the one or more trained neural networks;
• and
• provide the identified compression profile.

A method, comprising:

• performing, by one or more computing devices:
• automatically searching for a compression profile that optimizes a performance characteristic of one or more neural networks trained according to a respective one or more data sets, wherein the searching comprises:
• generating different prospective compression profiles for the one or more trained neural networks;
• collecting performance information of the different prospective compression profiles for respectively compressed versions the one or more trained neural networks using the one or more data sets;
• evaluating the performance information of the different prospective compression profiles for the respective compressed versions of the one or more trained neural networks to identify one of the different prospective compression profiles that optimizes the performance characteristic of the one or more trained neural networks;
• and
• providing the identified compression profile.

One or more non-transitory computer-readable storage media storing program instructions that, when executed on or across one or more computing devices, cause the one or more computing devices to implement:

• automatically searching for a compression profile that optimizes a performance characteristic of one or more neural networks trained according to a respective one or more data sets, wherein the searching comprises:
• generating different prospective compression profiles for the one or more trained neural networks;
• collecting performance information of the different prospective compression profiles for respectively compressed versions the one or more trained neural networks using the one or more data sets;
• evaluating the performance information of the different prospective compression profiles for the respective compressed versions of the one or more trained neural networks to identify one of the different prospective compression profiles that optimizes the performance characteristic of the one or more trained neural networks;
• and
• providing the identified compression profile.