One of the gems dug up by Deep Mind in the field of neural learning is Neural Turing Machines. By extending the capabilities of a simple neural network by attaching a memory to it, neural networks are able to learn simple algorithms such as copying sorting and etc. The project involves implementing a Neural Turing Machine in the current ANN framework and testing it with the preliminary algorithms.
In addition, I will also implement batch normalization and GRU cell as part of my project. Batch normalization is an effective technique which has become an essential component in stabilizing the deep network learning. GRU cells are essentially simple forms of LSTM cell, yet for some tasks have been proven to converge faster. Also, GRU cells have lesser parameters, making them more prone to overfitting and stable towrds training.