Fecha de publicación: -- Fuente: WIPO "tomato" Various implementations relating to a system and method for determining attribution data for a training data instance using a secondary machine learning model are disclosed herein. In one implementation, a method may include receiving attribution query data for a training data instance with respect to a model output provided by a primary machine learning model. The method may also include identifying candidate training data, where the primary machine learning was previously trained using the candidate training data. The method may further include using a secondary machine learning model to determine attribution data for the training data instance, which may include: determining an adjacency matrix based on the candidate training data and determining the attribution data based on the adjacency matrix. The method may additionally include generating output data for the training data instance based on the attribution data.