Inference verification of machine learning algorithms

摘要

PROBLEM TO BE SOLVED: To provide an inference verification method for verifying a trained first machine learning algorithm. A set of data samples is input to each of a plurality of different trained machine learning algorithms, and a set of results is obtained from each algorithm. Multiple trained machine learning algorithms are trained using training data samples where at least some of the results differ from the training data samples used to train the first algorithm. It is the same as the first algorithm to be verified. For each sample in the dataset entered into multiple trained machine learning algorithms, it is determined whether all the results from the multiple trained machine learning algorithms are the same. If all the results are the same, the first algorithm is reported as potentially defective for that sample in the input dataset. [Selection diagram] FIG. 1A