METHOD AND APPARATUS FOR CONTINUAL FEW-SHOT LEARNING WITHOUT FORGETTING

摘要

An apparatus and method for continuous pew-shot running are disclosed. A model for the basic task is created with default classification weights for the base class of the basic task. A series of new jobs are received sequentially. Upon receiving each new task in the series of new tasks, the model is updated with a new classification weight for the new class of each new task. the new classification weight is based on one or more of the default classification weights, and when one or more other new tasks have previously been received in the series of new tasks, one or more other new classification weights for a new class of the one or more other new tasks Based on , it is generated by the weight generator. Further, for each new task, a first set of samples of each new task is classified into the new class using the updated model.