Alex Graves

A Neural Network with Memory. Recurrent Neural Networks have subastantial ability to remember. memry tied to [[Latent Space]]. Cant increase memory without size of NN.

Content is held in latent space. One step further by removing it from activation and moving to external.

Train everyting end to end with Gradient descent

programs have discrete nature, and gradient descent is continuous.

Key : how network interacts with memoery.

Neural training machine.

Content, memero allocation and temporal order.

It was able to read and interpret graph structures.

simple models tend to work well