Detecting model inversion attacks in federated learning

A vehicle comprising:

• a processing system, the processing system comprising:
• a memory comprising computer readable instructions;
• and
• a processing device for executing the computer readable instructions, the computer readable instructions controlling the processing device to perform operations comprising:
• receiving, by the processing system from a coordinating server, an initial machine learning model to be trained by the processing system at the vehicle using data associated with the vehicle to generate a trained local machine learning model, the trained local machine learning model being one of a plurality of local machine learning models that the coordinating server uses to generate a global machine learning model using federated learning;
• analyzing, by the processing system, the initial machine learning model to determine whether the initial machine learning model is a compromised model prior to training the initial machine learning model;
• responsive to determining that the initial machine learning model is a compromised model, implementing a corrective action to prevent training the initial machine learning model using the compromised model;
• responsive to determining that the initial machine learning model is not a compromised model, training, by the processing system, the initial machine learning model to generate the trained local machine learning model using data associated with the processing system;
• and
• transmitting a gradient associated with the trained local machine learning model to the coordinating server.

A computer program product comprising a computer readable storage medium having program instructions embodied therewith, the program instructions executable by at least one processor to cause the at least one processor to perform operations comprising:

• receiving, from a coordinating server, an initial machine learning model to be trained by the at least one processor using data associated with the at least one processor to generate a trained local machine learning model, the trained local machine learning model being one of a plurality of local machine learning models that the coordinating server uses to generate a global machine learning model using federated learning;
• analyzing the initial machine learning model to determine whether the initial machine learning model is a compromised model prior to training the initial machine learning model, wherein analyzing the initial machine learning model comprises:
• for each data item in a training batch, computing a gradient of a cost function at the data item with respect to a parameter;
• responsive to the gradient being non-zero, incrementing a counter;
• and
• comparing the counter to a threshold to determine whether the counter satisfies a threshold;
• and
• responsive to determining that the initial machine learning model is a compromised model, implementing a corrective action to prevent training the initial machine learning model using the compromised model.