Mathematical Biology Seminar Andy Liu, University of Utah, Tuesday, Sep. 17, 2024 2:00pm in LCB 323 Deep reinforcement learning trains agents to develop useful internal representations. Abstract: Systems neuroscience has seen the development of many useful tools to study not just what the brain does, but how it can learn to solve complex tasks. We ask questions about what representations brains have available - information about the world encoded within the activity patterns of their neurons - to make decisions with. It turns out that when you match general-purpose learning neural networks with reinforcement learning, you get agents with representations that look a lot like ones in the brain. Having access to those artificial brains opens up all sorts of interesting things for us to explore.