US 12,651,480 B2 · Data set generation and augmentation for machine learning models

Abstract

A machine learning model (MLM) may be trained and evaluated. Attribute-based performance metrics may be analyzed to identify attributes for which the MLM is performing below a threshold when each are present in a sample. A generative neural network (GNN) may be used to generate samples including compositions of the attributes, and the samples may be used to augment the data used to train the MLM. This may be repeated until one or more criteria are satisfied. In various examples, a temporal sequence of data items, such as frames of a video, may be generated which may form samples of the data set. Sets of attribute values may be determined based on one or more temporal scenarios to be represented in the data set, and one or more GNNs may be used to generate the sequence to depict information corresponding to the attribute values.

Claims20

Dependent claims

A method comprising:

obtaining output data indicating predictions made by at least one machine learning model (MLM), the predictions corresponding to one or more objects represented in one or more data samples applied to the at least one MLM, and being based at least on compositions of attributes of the one or more objects;
evaluating the output data to determine one or more performance metric values corresponding to performance of the at least one MLM with respect to the predictions for a subset of the attributes;
computing one or more values representing a numerical quantity of additional data samples corresponding to the subset of the attributes, wherein the numerical quantity increases with one or more distances between the one or more performance metric values and one or more corresponding threshold values;
applying at least one input to one or more generative MLMs to generate, using the one or more values, the additional data samples;
and
updating parameters comprised in the at least one MLM using the additional data samples.

3 more elements

A system comprising:

one or more processing units to execute operations comprising:
obtaining output data indicating predictions made by at least one machine learning model (MLM), the predictions corresponding to one or more objects depicted in one or more first images applied to the at least one MLM, and being based at least on compositions of attributes of the one or more objects;
generating, using the output data, one or more performance metric values corresponding to performance of the at least one MLM with respect to the predictions for a subset of the attributes;
computing, using the one or more performance metric values, one or more values representing a numerical quantity of images corresponding to the subset of the attributes, wherein the numerical quantity increases with one or more distances between the one or more performance metric values and one or more corresponding threshold values;
applying at least one input to one or more generative MLMs to generate, using the one or more values, the images;
and
updating at least one parameter comprised in the at least one MLM using at least the images.

4 more elements

Dependencies are inferred from the claim text.

Data set generation and augmentation for machine learning models

Life of this patent

Data set generation and augmentation for machine learning models

Life of this patent

Abstract

Claims20