Abstract—Despite many successful efforts have been made in one-shot/few-shot learning tasks recently, learning from few data remains one of the toughest areas in machine learning. Regarding traditional machine learning methods, the more data gathered, the more accurate the intelligent system could perform. Hence for this kind of tasks, there must be different approaches to guarantee systems could learn even with only one sample per class but provide the most accurate results. Existing solutions are ranged from Bayesian approaches to meta-learning approaches. With the advances in Deep learning field, meta-learning methods with deep networks could be better such as recurrent networks with enhanced external memory (Neural Turing Machines - NTM), metric learning with Matching Network framework, etc. In our research, we propose a metric learning method for few-shot learning tasks by taking advantage of NTMs and Matching Network to improve few-shot learning task’s learning accuracy on both Omniglot dataset and Mini-Imagenet dataset. In addition, a weighted prototype is also introduced to improve the overall performance of the proposed model, especially on the complicated benchmark datasets such as mini-ImageNet.
Index Terms—Few shot learning, matching network, memory augmented neural network, prototypical network.
The authors are with the Department of Computer Science at National Defense Academy of Japan (e-mail: t2kien@ gmail.com, hsato@nda.ac.jp, masaok@nda.ac.jp).
Cite: Kien Tran, Hiroshi Sato, and Masao Kubo, "Memory Augmented Matching Networks for Few-Shot Learnings," International Journal of Machine Learning and Computing vol. 9, no. 6, pp. 743-748, 2019.
Copyright © 2019 by the authors. This is an open access article distributed under the Creative Commons Attribution License which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited (CC BY 4.0).