International Business Machines Corporation

University of Michigan

Mikhail Yurochkin

Sohini Upadhyay

Yasaman Khazaeni

Mayank Agarwal

Yuekai Sun

Djallel Bouneffouf

Trong Nghia Hoang

Soumya Ghosh

Debarghya Mukherjee

Moulinath Banerjee

Amanda Ruth Garcia Bower

Seyed Hamid Eftekhari

Alexander Vargo

Kristjan Herbert Greenewald

Fan Zhang

Obtain a first data set, a second data set, and a machine learning model. Construct a sensitive subspace of the first data set that defines a fair metric for distance among elements of the first data set. Fairly train the machine learning model on the first data set using a distributionally robust optimization approach based on the fair metric. Produce an individually fair set of labels by applying the fairly trained machine learning model to the second data set. Allocate a resource according to the individually fair set of labels.

Training individually fair machine learning algorithms via distributionally robust optimization

The present invention provides techniques for learning Mahalanobis distance similarity metrics from data for individually fair machine learning models. In one aspect, a method for learning a fair Mahalanobis distance similarity metric includes: obtaining data with similarity annotations; selecting, based on the data obtained, a model for learning a Mahalanobis covariance matrix Σ; and learning the Mahalanobis covariance matrix Σ from the data using the model selected, wherein the Mahalanobis covariance matrix Σ fully defines the fair Mahalanobis distance similarity metric.

Learning Mahalanobis distance metrics from data

In federated learning problems, data is scattered across different servers and exchanging or pooling it is often impractical or prohibited. A Bayesian nonparametric framework is presented for federated learning with neural networks. Each data server is assumed to provide local neural network weights, which are modeled through our framework. An inference approach is presented that allows us to synthesize a more expressive global network without additional supervision, data pooling and with as few as a single communication round. The efficacy of the present invention on federated learning problems simulated from two popular image classification datasets is shown.

Bayesian nonparametric learning of neural networks

A feature vector characterizing a system to be analyzed via online partially rewarded machine learning is obtained. Based on the feature vector, a decision is made, via the machine learning, using an online policy. The system is observed for environmental feedback. In at least a first instance, wherein the observing indicates that the environmental feedback is available, the environmental feedback is obtained. In at least a second instance, wherein the observing indicates that the environmental feedback is missing, the environmental feedback is imputed via an online imputation method. the online policy is updated based on results of the obtained environmental feedback and the online imputation method. A decision is output based on the updated online policy.

Online partially rewarded learning

Growing community of inventors

Cambridge, MA, United States of America

Mikhail Yurochkin