Systems and methods for generating and applying a secure statistical classifier

A system for secure classification of input, comprising:

• at least one hardware processor executing a code for:
• receiving input for classification by a secure statistical classifier;
• providing the input to each of a plurality of trained sub-classifiers, wherein each of the plurality of trained sub-classifiers includes a respective instance of an untrained statistical classifier trained on a sub-training dataset of a plurality of sub-training datasets, the each sub-training dataset including a plurality of records, each record including a combined input data item generated from a combination of a training input data item of a main training dataset and at least one unique cryptographic key of a plurality of cryptographic keys, and a ground truth label of the training input data item included in the combined input data item, by iteratively adjusting adjustable classification parameters of the respective instance of the untrained statistical classifier according to the combined input data item of the sub-training dataset serving as input generated from the combination of the training input data item and the at least one unique cryptographic key and the corresponding ground truth label,
• wherein the adjustable classification parameters of each trained sub-classifier have unique values computed according to the corresponding at least one unique cryptographic key of the sub-training dataset;
• and
• computing a single classification output in response to the plurality of classification outputs from the plurality of trained sub-classifiers.

A method of computing a secure statistical classifier, comprising:

• creating a plurality of sub-training datasets, each sub-training dataset including a plurality of records, each record including:
• a combined input data item generated from a combination of a training input data item of a main training dataset including a plurality of training input data items and a plurality of corresponding ground truth labels, and at least one unique cryptographic key of a plurality of cryptographic keys, and
• a ground truth label of the training input data item included in the combined input data item;
• creating a plurality of trained sub-classifiers by training each of a plurality of instances of an untrained statistical classifier on a sub-training dataset of the plurality of sub-training datasets by iteratively adjusting adjustable classification parameters of the respective instance of the untrained statistical classifier according to the combined input data item of the sub-training dataset serving as input generated from the combination of the training input data item and the at least one unique cryptographic key and the corresponding ground truth label;
• and
• providing the secure statistical classifier that includes the plurality of trained sub-classifiers.