Identifying contributing training datasets for outputs of machine learning models

A system comprising:

• one or more processors;
• a memory storing instructions which, when executed by the one or more processors, causes performance of:
• storing, in the memory, a plurality of training examples comprising training inputs and training outputs;
• generating a machine learning model and training the machine learning model using the plurality of training examples, wherein during the training of the machine learning model, each training example, of the plurality of training examples, has a corresponding initial weight;
• receiving a particular input for the machine learning model and, using the particular input and the machine learning model, computing a particular output;
• for each training example of the plurality of training examples:
• adjusting a weight of the training example to an adjusted weight that is different from the initial weight that corresponds to that training example;
• after adjusting the weight of the training example to the adjusted weight, generating a retrained machine learning model by retraining the machine learning model based on the training example with the adjusted weight;
• using the particular input and the retrained machine learning model, computing a second output;
• and
• based on a difference between the particular output and the second output, computing a relative numerical impact value on the particular output for the training example, wherein the relative numerical impact value reflects an importance of the training example on the particular output relative to an importance of the other training examples of the plurality of training examples on the particular output;
• generating training example relevance data comprising identifiers of the plurality of training examples and the relative numerical impact values for the plurality of training examples;
• storing the training example relevance data in the memory.