Professor John Kelleher discusses recurrent neural networks and conversational AI

Professor Kelleher talks about his interest in sequence prediction and long distance dependence in the context of NLP and notes that neural machine translation is a natural application of sequential data.. Professor Kelleher discusses why recurrent neural networks are particularly good at machine translation due to the sequential nature of language and allowing the system to have context. The encoder-decoder recurrent neural network architecture is the core technology inside Google's assistants for example. Thus employing recurrent neural network systems and other techniques will play a key role in building the next generation of dialog devices.