METHOD AND SYSTEM TO COLLABORATIVELY TRAIN DATA ANALYTICS MODEL PARAMETERS

A computer-implemented method, a computer program and a computer system for collaboratively training a model by a plurality of nodes in a computer network are provided. The method comprises receiving, by each of the nodes, the model and storing, by each of the nodes, corresponding local data for training the model. The method further comprises computing, by each of the nodes and according to an algorithm, parameter values of the model using the respective local data of the node and obfuscating, by a first node of the nodes, the computed parameter values of the first node. The method further comprises progressively training the obfuscated parameter values at each further node of the nodes, by adding the computed parameter values of the respective node to parameter values received from the previous node, whereby a final node of the nodes produces modified parameter values. The method further comprises receiving the modified parameter values at the first node and producing trained parameter values for the model by removing, by the first node, obfuscation of the modified parameter values. The method further comprises distributing, to all other nodes of the plurality of nodes, the trained parameter values.