Attentive Recurrent Comparators
This addresses the problem of rapid learning in AI for tasks like one-shot classification, representing a significant advance over prior methods.
The authors tackled one-shot classification on the Omniglot dataset by developing Attentive Recurrent Comparators (ARCs) to form object representations, achieving a state-of-the-art error rate of 1.5%, which is the first super-human result with a generic pixel-based model.
Rapid learning requires flexible representations to quickly adopt to new evidence. We develop a novel class of models called Attentive Recurrent Comparators (ARCs) that form representations of objects by cycling through them and making observations. Using the representations extracted by ARCs, we develop a way of approximating a \textit{dynamic representation space} and use it for one-shot learning. In the task of one-shot classification on the Omniglot dataset, we achieve the state of the art performance with an error rate of 1.5\%. This represents the first super-human result achieved for this task with a generic model that uses only pixel information.