Neural Turing Machines (Paper Explained)

*Overview*: Neural Networks (even RNNs, LSTMs etc) suffer from forgetting the input. NTM proposes to use external memory, similar to a CPU-Registers-RAM architecture. They propose novel ways of addressing the memory using end-to-end differentiable operations. *Authors*: Alex Graves, Greg Wayne, Ivo Danihelka *Abstract*: We extend the capabilities of neural networks by coupling them to external memory resources, which they can interact with by attentional processes. The com