Hands-On Artificial Intelligence for Beginners

Neural turing machines

Neural turing machines (NTM) are a form of RNN that were developed by Alex Graves of DeepMind in 2014. Instead of having an internal state such as an LSTM, NTMs have external memory, which increases their ability to handle complex computational tasks.

 NTMs consist of two main components: a controller and a memory bank.

• NTM controller: The controller in an NTM is the neural network itself; it manages the flow of information between input and memory, and learns to manage its own memory throughout
• NTM memory bank: The actual external memory, usually represented in tensor form

Provided there's enough memory, an NTM can replicate any algorithm by simply updating its own memory to reflect that algorithm. The architecture ...