AmazonGranted patentAI & softwareInterest 4/10

Amazon patents method to stop AI tuning when results are good enough

Amazon has been granted a patent for a method that automatically stops the process of tuning artificial intelligence models. The system calculates a measure of how much better the model could potentially be and halts the search once that potential improvement falls below a specific limit.

Why it mattersThis improves the efficiency of building AI models by preventing computers from wasting time searching for tiny improvements that are not worth the cost.

In the filing’s own words

Techniques for hyperparameter optimization are described. In particular, examples detailed herein describe stopping hyperparameter optimization based at least in part on a comparison of a threshold value to a simple regret value. In some examples, the threshold value is provided by a request to perform hyperparameter optimization. In some examples, the threshold value is calculated during hyperparameter optimization.

What exactly they claim (claim 1)

1. A computer-implemented method comprising: receiving, at a hyperparameter optimization (HPO) service implemented as HPO service code stored within memory and executed by one or more processors of the HPO service, a request to perform HPO to determine a value for at least one hyperparameter of a machine learning algorithm, the request including an indication that the HPO is to automatically terminate based at least in part on a threshold value compared to a simple regret value, wherein the request to perform HPO to determine the value for the at least one hyperparameter of the machine learning algorithm does not include the threshold value and the threshold value is determined by a cross-validation estimate of a generalization error; determining, by the HPO service according to the request, a hyperparameter value for the at least one hyperparameter of the machine learning algorithm using HPO based on an evaluation of one or more hyperparameters from the machine learning algorithm on different datasets, wherein the evaluation is to stop when a computed upper bound for simple regret is smaller than the threshold; and tuning the machine learning algorithm using the hyperparameter value for the at least one hyperparameter of the machine learning algorithm to generate a machine learning model.

Facts

Company
Amazon (Amazon Technologies, Inc.)
Official title
Automatic determination of hyperparameter optimization termination
Patent number
US12792973B1
Application number
17847046
Filed
Jun 22, 2022
Granted
Nov 10, 2026
Inventors
6
Claims
20

Classified under

G06N 20/20 Physics & computing

More from Amazon in AI & software

Summaries are simplified from the filing’s own abstract and first claim. A patent filing is not a product announcement.
Patent data and drawings: U.S. Patent and Trademark Office (USPTO). Nebula is not affiliated with or endorsed by the USPTO.