Systems and methods for active learning from sparse training data

Patent metadata

Field	Value
Publication number	US-11710038-B2
Application number	US-202016847098-A
Country	US
Kind code	B2
Filing date	Apr 13, 2020
Priority date	Apr 13, 2020
Publication date	Jul 25, 2023
Grant date	Jul 25, 2023

Abstract

A method for active learning using sparse training data can include training a machine learning model using less than ten first training data points to generate a candidate machine learning model. The method can include performing a Monte Carlo process to sample one or more first outputs of the candidate machine learning model. The method can include testing the one or more first outputs to determine if each of the one or more first outputs satisfy a respective convergence condition. The method can include, responsive to at least one first output not satisfying the respective convergence condition, training the candidate machine learning model using at least one second training data point corresponding to the at least one first output. The method can include, responsive to the one or more first outputs each satisfying the respective convergence condition, outputting the candidate machine learning model.

First claim

What is claimed is:

1. A method for machine learning using sparse training data, comprising:
training, by one or more processors, a machine learning model using a first training data point of a plurality of training data points to generate a trained machine learning model, wherein training the machine learning model comprises modifying the machine learning model to satisfy a first convergence condition;
selecting, by the one or more processors, a first input to provide to the trained machine learning model, the first input selected from a domain of inputs that includes inputs of the plurality of training data points by performing at least one of a random process or a Monte Carlo process, the first input associated with a first training data output;
sampling, by the one or more processors, a first candidate output of the trained machine learning model by providing the first input as input to the trained machine learning model;
determining, by the one or more processors, whether the first candidate output satisfies a second convergence condition based on a difference between the first training data output and the first candidate output, the second convergence condition different from the first convergence condition;
responsive to the first candidate output not satisfying the second convergence condition, modifying, by the one or more processors, the trained machine learning model using a second training data point of the plurality of training data points, the second training data point comprising the first input and the first training data output; and
“responsive to the first candidate output satisfying the second convergence condition, outputting, by the one or more processors, the trained machine learning model.

2. The method of claim 1 , wherein the first training data point is retrieved from a dataset comprising less than one thousand training data points.

3. The method of claim 1 , wherein the machine learning model comprises a neural network comprising an input layer, one or more hidden layers, and an output layer.

4. The method of claim 1 , wherein the first training data output is a reference output of a reference high-fidelity model.

5. The method of claim 1 , wherein training the machine learning model using the first training data point comprises retraining the machine learning model using less than ten training data points.

6. The method of claim 1 , wherein the first training data point comprises at least one of an experimental data point or a synthetic data point from a reference high-fidelity model.

7. The method of claim 1 , wherein the plurality of training data points are from a first subset of data points of a training database, and the domain of inputs comprises a second subset of data points of the training database.

8. The method of claim 1 , wherein:
sampling, by the one or more processors, the first candidate output comprises sampling a plurality of first candidate outputs of the trained machine learning model;
determining, by the one or more processors, whether the first candidate output satisfies the second convergence condition comprises determining, by the one or more processors, whether each first candidate output of the plurality of first candidate outputs satisfies a respective second convergence condition; and
modifying, by the one or more processors, the trained machine learning model comprises retraining the trained machine learning model using one or more second training data points corresponding to one or more first candidate outputs of the plurality of first candidate outputs that did not satisfy the respective second convergence condition.

9. The method of claim 1 , wherein training the machine learning model comprises causing weights and biases of a neural network of the machine learning model to be adjusted until an optimization condition is satisfied.

10. A system, comprising:
one or more hardware processors configured to:
train a machine learning model using a first training data point of a plurality of training data points to generate a trained machine learning model, wherein training the machine learning model comprises modifying the machine learning model to satisfy a first convergence condition;
select a first input to provide to the trained machine learning model, the first input selected from a domain of inputs that includes inputs of the plurality of training data points by performing at least one of a random process or a Monte Carlo process, the first input associated with a first training data output;
sample a first candidate output of the trained machine learning model by providing the first input as input to the trained machine learning model;
determine whether the first candidate output satisfies a second convergence condition based on a difference between the first training data output and the first candidate output, the second convergence condition different from the first convergence condition;
modify, responsive to the first candidate output not satisfying the second convergence condition, the trained machine learning model using a second training data point of the plurality of training data points, the second training data point comprising the first input and the first training data output; and
output, responsive to the first candidate output satisfying the second convergence condition, the trained machine learning model.

11. The system of claim 10 , wherein the first training data point is retrieved from a training database comprising less than one thousand training data points.

12. The system of claim 10 , wherein the machine learning model comprises a neural network comprising an input layer, one or more hidden layers, and an output layer.

13. The system of claim 10 , wherein the one or more processors are configured to train the machine learning model using the first training data point by training the machine learning model using less than ten training data points.

14. The system of claim 10 , wherein the first training data point comprises at least one of an experimental data point or a synthetic data point from a reference high-fidelity model.

15. A method, comprising:
training a machine learning model using less than ten first training data points of a plurality of training data points to generate a trained machine learning model that satisfies a first convergence condition;
performing a Monte Carlo process to select a plurality of first inputs from the plurality of training data points, each first input of the plurality of first inputs corresponding to a first training data output of a plurality of first training data outputs;
applying the plurality of first inputs as input to the trained machine learning model to sample a plurality of first candidate outputs of the trained machine learning model;
testing the plurality of first candidate outputs to determine if each of the plurality of first candidate outputs satisfies a respective second convergence condition based on differences between the plurality of first candidate outputs and the plurality of first training data outputs, the respective second convergence condition different from the first convergence condition;
responsive to at least one first candidate output not satisfying the respective second convergence condition, training the trained machine learning model using at least one second training data point of the plurality of training data points corresponding to the at least one first candidate output, the at least one second training data point comprising at least one first input used as input to the trained machine learning model to sample the at least one first candidate output

Assignees

• Uchicago Argonne Llc

Inventors

• Sankaranarayanan Subramanian
• Loeffler Troy David
• Chan Henry

Classifications

• G06N3/0499

  Feedforward networks · CPC title

• G06N3/091

  Active learning · CPC title

• G06N3/09

  Supervised learning · CPC title

• G06N3/08 Primary

  Learning methods · CPC title

• G06N3/04

  Architecture, e.g. interconnection topology · CPC title