How Do Deep Neural Networks Learn to Learn? A Neuroscientist's Perspective
Despite the impressive successes of deep reinforcement learning in a wide variety of domains, such approaches still require orders of magnitude more data than a human needs to learn. In this talk, I'll explore the possible reasons for this difference, why humans are so much better at learning, and efforts in artificial intelligence to close this gap via meta-learning, or learning to learn.
Jane is a senior research scientist at DeepMind, where, as a member of the neuroscience team, she works to create new models of AI and deep reinforcement learning, inspired by the latest advancements in cognitive neuroscience. She obtained her Ph.D in Applied Physics from the University of Michigan, where she studied complex systems, physics, and computational neuroscience, before moving on to conduct research in cognitive neuroscience at Northwestern University. She also enjoys science fiction and aerial arts.