When people learn, we can learn by being directly taught, as in school. But often, we learn by doing things, seeing what works and what doesn't, and learning a good strategy ourselves over time. This is the key insight behind reinforcement learning! Reinforcement learning is a machine learning paradigm where, rather than telling the program what to do, it is given a reward for good actions, and learns how to maximise that reward. It's the approach behind many of the biggest recent achievements in machine learning, such as mastering Go and StarCraft.

In this talk, Neel will explore the key underlying ideas, the key algorithms, and give an overview of some more modern applications using neural networks. He'll also explore what thinking about reinforcement learning can teach us about human decision-making, problems with optimisation for an imperfect reward, and the value of information.

This talk will focus on the high-level ideas and intuitions behind Reinforcement Learning, rather than the technical details. There are no pre-requisites.

Resources (including slides and references): https://www.neelnanda.io/rl-talk
Feedback form: https://forms.gle/KscasfzfF2y6trpf7

--------------------------------------------------------------------------------
Subscribe:    / @thearchimedeans7041  
Our website: https://archim.soc.srcf.net/
Our Facebook page:   / archimedeans  
Join our mailing list: https://lists.cam.ac.uk/mailman/listi...
Join our Microsoft Teams: http://forms.gle/tGHnN7XhEUVCeRgr6