Artificial Intelligence, and its diverse subfields, has been the subject of intense study for more than half a century. Recent advances in machine learning, jointly known as deep learning, have partially closed the gap that exists between the abilities of naturally intelligent systems (i.e., brains) and artificially intelligent ones, particularly in problems related with perception.
Some problems that have been deemed very hard to solve, like learning to play the game of Go or other complex strategy games, from scratch, have fallen to approaches that combine deep reinforcement learning with efficient computation methods.
However, the depth of understanding of these systems is still very limited, and DCNNs (deep convolutional neural networks) are still very far from approaching human abilities even in simple perception problems. In this seminar, I will lead a discussion about the power and limitations of DCNNs, with practical examples and challenges.
