US 11,775,863 B2 · Enforcing fairness on unlabeled data to improve modeling performance

Abstract

Fairness of a trained classifier may be ensured by generating a data set for training, the data set generated using input data points of a feature space including multiple dimensions and according to different parameters including an amount of label bias, a control for discrepancy between rarity of features, and an amount of selection bias. Unlabeled data points of the input data comprising unobserved ground truths are labeled according to the amount of label bias and the input data sampled according to the amount of selection bias and the control for the discrepancy between the rarity of features. The classifier is then trained using the sampled and labeled data points as well as additional unlabeled data points. The trained classifier is then usable to determine unbiased classifications of one or more labels for one or more other data sets.

Claims19

Dependent claims

A system, comprising:

at least one processor;
a memory, comprising program instructions that when executed by the at least one processor cause the at least one processor to implement a machine learning system configured to train an unbiased classifier, the unbiased classifier configured to determine unbiased classifications of one or more labels for the one or more data sets, wherein to generate the unbiased classifier the machine learning system is configured to:
generate a training data set from samples of input data points of a feature space comprising a plurality of dimensions according to a plurality of training parameters, wherein the training data set comprises an amount of label bias and an amount of selection bias, and wherein the training parameters comprise:
a value indicating the amount of label bias;
another value indicating the amount of selection bias;
and
a control for discrepancy between rarity of features;
and
train the unbiased classifier using the generated training data set and the plurality of training parameters.

6 more elements

One or more non-transitory computer-accessible storage media storing program instructions that when executed on or across one or more processors cause the one or more processors to implement a machine learning system performing:

training an unbiased classifier that, when applied to one or more data sets determines unbiased classifications of one or more labels for the one or more data sets, the training comprising:
generating a training data set from samples of input data of a feature space comprising a plurality of dimensions according to a plurality of training parameters, wherein the training data set comprises an amount of label bias and an amount of selection bias, and wherein the training parameters comprise:
a value indicating the amount of label bias;
another value indicating the amount of selection bias;
and
a control for discrepancy between rarity of features;
and
training the unbiased classifier using the generated training data set.

5 more elements

Dependencies are inferred from the claim text.

Enforcing fairness on unlabeled data to improve modeling performance

Life of this patent

Enforcing fairness on unlabeled data to improve modeling performance

Life of this patent

Abstract

Claims19