nsokhand / how-to-learn-from-little-data

This is the code for "How to Learn from Little Data - Intro to Deep Learning #17' by Siraj Raval on YouTube

One Shot Learning using Memory-Augmented Neural Networks in Tensorflow.

Coding Challenge - Due Date, Thursday May 11th 2017

This weeks coding challenge is to use this code to classify 2 animals! You can use this dataset. That's it. No one has ever tried to apply memory augmented networks to anything but the omniglot dataset, so just a simple attempt at this would be both an awesome learning exercise and hugely useful to the AI community. Document your results in your README, regardless of whether the accuracy is high or not. Good luck!

Overview

This is the code for this video on Youtube by Siraj Raval as part of the Udacity Deep Learning nanodegree. We're going to use a Memory Augmented Neural Network in Tensorflow to perform one-shot learning on the omniglot dataset. That means we'll be able to classify images by their labels after training on just a few samples.

The benchmark dataset is Omniglot dataset. All the datasets should be placed in the data/ folder.

Dependencies

• tensorflow (Written for Tensorflow v0.12. Yet to add support for Tensorflow v1*.)
• numpy
• time

Install dependencies using pip

Usage

run python omniglot.py in terminal to train the model after installing the necessary dependencies.

Credits

The credits for this code go to hmishra2250 i've merely created a wrapper to get people started.