资源算法ntm-meta-learning

ntm-meta-learning

2020-02-26 | |  46 |   0 |   0

Meta-Learning with Memory Augmented Neural Networks

A chainer implementation of Meta-Learning with Memory Augmented Neural Networks
(This paper is also known as One-shot Learning with Memory Augmented Neural Networks )

  • Adam Santoro, Sergey Bartunov, Matthew Botvinick, Daan Wierstra, Timothy Lillicrap, Meta-Learning with Memory-Augmented Neural Networks, [link]

  • Some code is taken from tristandeleu's implementation with Lasagne.

How to run

  1. Download the Omniglot dataset and place it in the data/ folder.

  2. Run the scripts in data/omniglot to prepare dataset.

  3. Run scripts/train_omniglot.py (Use gpu option if needed)

Summary of the paper

The authors attack the problem of one-shot learning by the approach of meta-learning. They propose Memory Augmented Neural Network, which is a variant of Neural Turing Machine, and train it to learn "how to memorize unseen characters." After the training, the model can learn unseen characters in a few shot.


上一篇:ntm_keras

下一篇:NTMS-web

用户评价
全部评价

热门资源

  • TensorFlow-Course

    This repository aims to provide simple and read...

  • seetafaceJNI

    项目介绍 基于中科院seetaface2进行封装的JAVA...

  • mxnet_VanillaCNN

    This is a mxnet implementation of the Vanilla C...

  • vsepp_tensorflow

    Improving Visual-Semantic Embeddings with Hard ...

  • DuReader_QANet_BiDAF

    Machine Reading Comprehension on DuReader Usin...