Another amazing year of GSOC. This year was more exciting that my previous experiences.
My major contribution in this year, was the Neural Turing Machine, or NTM. NTM is described in this. NTM is a neural network with additional external memory. As the entire architecture is end-to-end differentiable, NTM is able to learn tasks which require memory storage. Such as, reproducing the given sequence, sorting the sequence and so on. The most complex part of NTM is the MemoryHead. MemoryHead is the module which tells the neural network which memory location to access, based on the memory content.