Last week, I attended the Deep Learning Summit in Montreal organized by RE.WORK. Despite of my expectation to see a main population of NLP and Vision scientists, a lot of attendees came from other areas. It does sound like Deep Neural Networks are being applied to many problems that are only remotely related to human cognition. Using distributional semantic representations as heuristics for compiling computer programs was one of such interesting topics, in which the classic fully algorithmic/discrete techniques were combined with differential neural network computations.
A very interesting part of the program was the panel discussion among the three Deep Learning pioneers: Hinton, LaCun and Bengio. I recommend reading the interview here. Each of the giants had a separate talk as well, where they presented some ideas from their current research. After Bengio’s talk I approached him for a little chat on my project and possible solutions for fake news. He believes, with currently available data, solving the fake news problem with a Deep NLP approach sounds hopeless!