Methods and systems for secure data analysis and machine learning

A method for secure distributed machine learning, the method comprising, by a processor:

• receiving a request for training a machine learning model at a plurality of training devices;
• receiving a labeled dataset for use in training the machine learning model;
• transmitting, to the plurality of training device configured for training the machine learning model, a plurality of clusters of training data selected from the labeled dataset, wherein each of the plurality of clusters has less than a threshold amount of data that is determined to prevent a training device from deriving information about the labeled dataset from that cluster;
• receiving, from each of the plurality of training devices, a partially trained machine learning model;
• for each of the plurality of the partially trained machine learning models:
• evaluating that partially trained machine learning model to determine whether that partially trained machine learning model satisfies an evaluation criterion,
• in response to determining that the partially trained machine learning model does not satisfy the evaluation criterion, transmitting to a training device from which that partially trained machine learning model was received, a second cluster of training data selected from the labeled dataset for use in updating the partially trained machine learning model, wherein the second cluster has less than the threshold amount of data, and
• receiving an updated partially trained machine learning model.

A system for secure machine learning, the system comprising:

• a processor;
• and
• a non-transitory computer readable medium comprising one or more programming instructions that when executed by the processor, cause the processor to:
• receive a labeled dataset for use in training the machine learning model;
• transmit, to a training device, a first cluster of training data selected from the labeled dataset, wherein the first cluster has less than a threshold amount of data that is determined to prevent the training device from deriving information about the labeled dataset from the first cluster;
• receive, from the training device, a trained machine learning model;
• evaluate the trained machine learning to determine whether the trained machine learning model satisfies an evaluation criterion;
• in response to determining that the trained machine learning model satisfies the evaluation criterion, encrypt the trained machine learning model;
• and
• deploy the encrypted machine learning model to make predictions using live data.

A system for secure distributed machine learning, the method comprising, by a processor:

• a processor;
• and
• a non-transitory computer readable medium comprising one or more programming instructions that when executed by the processor, cause the processor to:
• receive a request for training a machine learning model at a plurality of training devices;
• receive a labeled dataset for use in training the machine learning model;
• transmit, to the plurality of training device configured for training the machine learning model, a plurality of clusters of training data selected from the labeled dataset, wherein each of the plurality of clusters has less than a threshold amount of data that is determined to prevent a training device from deriving information about the labeled dataset from that cluster;
• receive, from each of the plurality of training devices, a partially trained machine learning model;
• for each of the plurality of the partially trained machine learning models:
• evaluate that partially trained machine learning model to determine whether that partially trained machine learning model satisfies an evaluation criterion,
• in response to determining that the partially trained machine learning model does not satisfy the evaluation criterion, transmit to a training device from which that partially trained machine learning model was received, a second cluster of training data selected from the labeled dataset for use in updating the partially trained machine learning model, wherein the second cluster has less than the threshold amount of data, and
• receive an updated partially trained machine learning model.