Systems and methods for an accelerated tuning of hyperparameters of a model using a machine learning-based tuning service

Patent metadata

Field	Value
Publication number	US-12373699-B2
Application number	US-202318320758-A
Country	US
Kind code	B2
Filing date	May 19, 2023
Priority date	Jul 13, 2018
Publication date	Jul 29, 2025
Grant date	Jul 29, 2025

Abstract

A system and method for accelerated tuning of hyperparameters includes receiving a multi-task tuning work request for tuning hyperparameters of a model, wherein the multi-task tuning work request includes: a full tuning task for tuning hyperparameters, wherein the full tuning task includes a first set of tuning parameters governing a first tuning operation; a partial tuning task for tuning the hyperparameters of the model, wherein the partial tuning task includes a second distinct set of tuning parameters governing a second tuning operation; executing the first tuning operation and the second tuning operation; generating a first suggestion set and a second suggestion set of one or more proposed values for the hyperparameters based on the execution of the full tuning task and the partial tuning task; and setting the partial tuning task as a proxy for the full tuning task thereby accelerating a tuning of the hyperparameters of the model.

First claim

What is claimed is:

1. A system comprising:
memory;
machine-readable instructions; and
programmable circuitry, the machine-readable instructions to cause the programmable circuitry to:
access a tuning work request to tune hyperparameters of a model, the tuning work request to include:
a full tuning task and a corresponding first number of datapoints;
a partial tuning task and a corresponding second number of datapoints, the second number of datapoints being less than the first number of datapoints;

execute a full tuning of the model based on the full tuning task;
execute a partial tuning of the model based on the partial tuning task;
evaluate a measured performance associated with the partial tuning of the model and a measured performance associated with the full tuning of the model;
select hyperparameter values associated with the partial tuning of the model based on a determination that the measured performance associated with the partial tuning of the model satisfies a performance value;
determine an importance of a first one of the hyperparameters of the model; and
based on the importance of the first hyperparameter, rank the first hyperparameter relative to other ones of the hyperparameters of the model.

2. The system of claim 1 , wherein the programmable circuitry is to, based on a budget, determine to use fewer than the second number of datapoints for the partial tuning of the model.

3. The system of claim 2 , wherein the programmable circuitry is to set a tuning cost parameter for the partial tuning of the model.

4. The system of claim 1 , wherein the first number of datapoints for the full tuning task is at least 1000 samples, and the second number of datapoints for the partial tuning task is at least 500 samples.

5. The system of claim 1 , wherein the model is a Latin Hypercube Sampling (LHS) model.

6. The system of claim 5 , wherein the partial tuning refers to a number of interchanges.

7. The system of claim 1 , wherein the model performs a dimensionality reduction method.

8. At least one of a random access memory (RAM), a read only memory (ROM), a flash memory, an electrically erasable programmable read only memory (EEPROM), an optical disc, or a hard drive comprising instructions to cause programmable circuitry to at least:
access a tuning work request to tune hyperparameters of a model, the tuning work request to include:
a full tuning task and a corresponding first number of datapoints;
a partial tuning task and a corresponding second number of datapoints, the second number of datapoints being less than the first number of datapoints;

execute a full tuning of the model based on the full tuning task;
execute a partial tuning of the model based on the partial tuning task;
evaluate a measured performance associated with the partial tuning of the model and a measured performance associated with the full tuning of the model;
select hyperparameter values associated with the partial tuning of the model based on a determination that the measured performance associated with the partial tuning of the model satisfies a performance value;
determine an importance of a first one of the hyperparameters of the model; and
prioritize the first hyperparameter relative to other ones of the hyperparameters of the model based on the importance of the first hyperparameter.

9. The at least one of the RAM, the ROM, the flash memory, the EEPROM, the optical disc, or the hard drive of claim 8 , wherein the instructions are to cause the programmable circuitry to, based on a budget, determine to use fewer than the second number of datapoints for the partial tuning of the model.

10. The at least one of the RAM, the ROM, the flash memory, the EEPROM, the optical disc, or the hard drive of claim 9 , wherein the instructions are to cause the programmable circuitry to set a tuning cost parameter for the partial tuning of the model.

11. The at least one of the RAM, the ROM, the flash memory, the EEPROM, the optical disc, or the hard drive of claim 8 , wherein the first number of datapoints for the full tuning task is at least 1000 samples, and the second number of datapoints for the partial tuning task is at least 500 samples.

12. The at least one of the RAM, the ROM, the flash memory, the EEPROM, the optical disc, or the hard drive of claim 8 , wherein the model is a Latin Hypercube Sampling (LHS) model.

13. The at least one of the RAM, the ROM, the flash memory, the EEPROM, the optical disc, or the hard drive of claim 12 , wherein the partial tuning refers to a number of interchanges.

14. The at least one of the RAM, the ROM, the flash memory, the EEPROM, the optical disc, or the hard drive of claim 8 , wherein the model performs a dimensionality reduction method.

15. A method to tune hyperparameters of a model based on a tuning work request, the method comprising:
accessing the tuning work request to tune the hyperparameters of the model, the tuning work request to include:
a full tuning task and a corresponding first number of datapoints;
a partial tuning task and a corresponding second number of datapoints, the second number of datapoints being less than the first number of datapoints;

executing, by executing an instruction with programmable circuitry, a full tuning of the model based on the full tuning task;
executing, by the programmable circuitry, a partial tuning of the model based on the partial tuning task;
evaluating, by the programmable circuitry, a measured performance associated with the partial tuning of the model and a measured performance associated with the full tuning of the model; and
selecting, by the programmable circuitry, hyperparameter values associated with the partial tuning of the model based on a determination that the measured performance associated with the partial tuning of the model satisfies a performance value;
determining an importance of a first one of the hyperparameters of the model; and
ranking the first hyperparameter relative to other ones of the hyperparameters of the model based on the importance of the first hyperparameter.

16. The method of claim 15 , including determining, based on a budget, to use fewer than the second number of datapoints for the partial tuning of the model.

17. The method of claim 16 , further including setting a tuning cost parameter for the partial tuning of the model.

18. The method of claim 15 , wherein the first number of datapoints for the full tuning task is at least 1000 samples, and the second number of datapoints for the partial tuning task is at least 500 samples.

19. The method of claim 15 , wherein the model is a Latin Hypercube Sampling (LHS) model.

20. The method of claim 19 , wherein the partial tuning refers to a number of interchanges.

Assignees

• Intel Corp

Inventors

• Mccourt Michael
• Hsu Ben
• Hayes Patrick
• Clark Scott

Classifications

• G06N20/00

  Machine learning · CPC title

• G06N20/20 Primary

  Ensemble learning · CPC title

• G06N7/01

  Probabilistic graphical models, e.g. probabilistic networks · CPC title

• G06N3/082 Primary

  modifying the architecture, e.g. adding, deleting or silencing nodes or connections · CPC title