Prior-knowledge and attention based meta-learning for few-shot learning

Recently, meta-learning has been shown to be a promising way solve few-shot learning. In this paper, inspired by the human cognition process, which utilizes both prior-knowledge and visual attention when learning new knowledge, we present novel paradigm of approach that capitalizes on three developments introduce mechanism meta-learning. our approach, is responsible for helping meta-learner express input data in high-level representation space, enables focus key features space. Compared with existing approaches pay little attention, alleviates meta-learner’s burden. Furthermore, discover Task-Over-Fitting (TOF) problem,1 indicates poor generalization across different K-shot tasks. To model TOF problem, propose Cross-Entropy Tasks (CET) metric.2 Extensive experiments demonstrate techniques improve state-of-the-art performance several benchmarks while also substantially alleviating problem.