Early experiment stopping for batch bayesian optimization in industrial processes

Patent metadata

Field	Value
Publication number	US-2022128972-A1
Application number	US-202017082663-A
Country	US
Kind code	A1
Filing date	Oct 28, 2020
Priority date	Oct 28, 2020
Publication date	Apr 28, 2022
Grant date	—

Abstract

Real-time intervention of an industrial process can include searching for a batch of candidate configurations for use by the industrial process, the batch of candidate configurations searched for by performing a batch Bayesian optimization (BBO). The batch of candidate configurations is transmitted to the industrial process to use in running the industrial process. A result of the run is received from the industrial process. Using the result in the BBO, a next batch of candidate configurations is searched. Whether a stopping criterion is met is determined, based on the next batch of candidate configurations and by applying a function to a BBO acquisition score. Responsive to determining that the stopping criterion is met, searching for the next batch of candidates is terminated.

First claim

What is claimed is:

1 . A computer-implemented method of real-time intervention of an industrial process, comprising:
searching for a batch of candidate configurations for use by the industrial process, the batch of candidate configurations searched for by performing a batch Bayesian optimization (BBO); transmitting the batch of candidate configurations to the industrial process to use in running the industrial process; receiving from the industrial process a result of the run; using the result in the BBO to search for a next batch of candidate configurations; determining whether a stopping criterion is met, based on the next batch of candidate configurations and by applying a function to a BBO acquisition score; responsive to determining that the stopping criterion is met, terminating a search for the next batch of candidates and controlling the industrial process to stop running; and responsive to determining that the stopping criterion is not met, transmitting the next batch of candidate configurations to the industrial process to use in running the industrial process and repeating the using the result in the BBO to search for a next batch of candidate configurations and the determining whether the stopping criterion is met.

2 . The method of claim 1 , wherein the function comprising evaluating data associated with the next batch of candidates based on a target criterion and a batch percentage criterion, the target criterion representing a statistical significance level needed to terminate the search, and the batch percentage criterion representing how much of a batch needs to fail to terminate the search.

3 . The method of claim 2 , wherein the determining whether a stopping criterion is met based on the next batch of candidate configurations includes:
for each of the candidate configurations in the batch, computing a contextual probability of improvement (cPI) score; and determining that a percentage of the candidates configurations with the cPI score less than the target criterion is greater than the batch percentage criterion.

4 . The method of claim 3 , wherein the cPI score is determined as a cumulative distribution function (CDF) of a standard normal distribution of a combination of a mean of variances contained within a sampled posterior distribution, a predicted value from a candidate and a best candidate value discovered among iterations of BBO.

5 . The method of claim 1 , further including:
responsive to determining that the stopping criterion is met, sending an optimal batch of configurations among batches of configurations found in the search to the industrial process.

6 . The method of claim 1 , further including:
responsive to determining that the stopping criterion is met, sending an optimal batch of configurations among batches of configurations found in the search to a user via a user interface.

7 . The method of claim 2 , wherein the target criterion and the batch percentage criterion are configurable.

8 . A system for real-time intervention of an industrial process, comprising:
a hardware processor; and a memory device coupled with the hardware processor; the hardware processor configured to at least:
search for a batch of candidate configurations for use by the industrial process, the batch of candidate configurations searched for by performing a batch Bayesian optimization (BBO);
transmit the batch of candidate configurations to the industrial process to use in running the industrial process;
receive from the industrial process a result of the run;
use the result in the BBO to search for a next batch of candidate configurations;
determine whether a stopping criterion is met, based on the next batch of candidate configurations and by applying a function to a BBO acquisition score;
responsive to determining that the stopping criterion is met, terminate a search for the next batch of candidates and control the industrial process to stop running; and
responsive to determining that the stopping criterion is not met, transmit the next batch of candidate configurations to the industrial process to use in running the industrial process and repeating using of the result in the BBO to search for a next batch of candidate configurations and determining of whether the stopping criterion is met.

9 . The system of claim 8 , wherein the hardware processor applying a function includes evaluating data associated with the next batch of candidates based on a target criterion and a batch percentage criterion, the target criterion representing a statistical significance level needed to terminate the search, and the batch percentage criterion representing how much of a batch needs to fail to terminate the search.

10 . The system of claim 9 , wherein to determine whether a stopping criterion is met based on the next batch of candidate configurations, the hardware processor is configured to:
for each of the candidate configurations in the batch, compute a contextual probability of improvement (cPI) score; and determine that a percentage of the candidates configurations with the cPI score less than the target criterion is greater than the batch percentage criterion.

11 . The system of claim 10 , wherein the cPI score is determined as a cumulative distribution function (CDF) of a standard normal distribution of a combination of a mean of variances contained within a sampled posterior distribution, a predicted value from a candidate and a best candidate value discovered among iterations of BBO.

12 . The system of claim 8 , wherein the hardware processor is further configured to:
responsive to determining that the stopping criterion is met, send an optimal batch of configurations among batches of configurations found in the search to the industrial process.

13 . The system of claim 8 , wherein the hardware processor is further configured to:
responsive to determining that the stopping criterion is met, send an optimal batch of configurations among batches of configurations found in the search to a user via a user interface.

14 . The system of claim 9 , wherein the target criterion and the batch percentage criterion are configurable.

15 . A computer program product comprising a computer readable storage medium having program instructions embodied therewith, the program instructions readable by a device to cause the device to:
search for a batch of candidate configurations for use by the industrial process, the batch of candidate configurations searched for by performing a batch Bayesian optimization (BBO); transmit the batch of candidate configurations to the industrial process to use in running the industrial process; receive from the industrial process a result of the run; use the result in the BBO to search for a next batch of candidate configurations; determine whether a stopping criterion is met, based on the next batch of candidate configurations and by applying a function to a BBO acquisition score; responsive to determining that the stopping criterion is met, terminate a search for the next batch of candidates, terminating the search controlling the industrial process to stop running; and responsive to determining that the stopping criterion is not met, transmit the next batch of candidate configurations to the industrial process to use in running the industrial process and repeat using of the result in the BBO to search for a next batch of candidate configurations and determining of whether the stopping criterion is met.

16 . The computer program product of claim 15 , wherein the device is caused to apply the function

Assignees

• IBM

Inventors

• Pyzer-Knapp Edward Oliver
• Fare Clyde

Classifications

• G06N7/01 Primary

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

• G06N3/063

  using electronic means · CPC title

• G06N20/00

  Machine learning · CPC title

• G05B19/4155 Primary

  characterised by program execution, i.e. part program or machine function execution, e.g. selection of a program · CPC title

• G05B2219/32396

  Job shop, batch production system · CPC title