Nunug

Topic: Reinforcement Learning

Presenter: Bruce Nielsen

Date: Thursday, January 10, 2019

Time: 6:30:00 PM -

Location:

Reinforcement Learning is a fascinating form of Machine Learning that allows an agent to learn from real world feedback over time rather than requiring labeled “desired outputs”, as is required in the more popular “Supervised Learning”. This makes Reinforcement Learning ideal for many types of tasks where we can’t reasonably come up with natural or expert labels/outputs. Growing up, I always heard that using Artificial Intelligence to beat a Chess master was easy (you just needed lots of memory with which to do a search) but we shouldn’t expect to see an AI beat a Go master within our lifetimes because of how complex the game of Go is. But Reinforcement Learning soon proved that prophecy false, when AlphaGo beat Lee Sedol, the world Go master, in 4 out of 5 games. Why was Reinforcement learning able to do this when other types of AI failed? Using Python code, mathematics, plus a bit of philosophy mingled with pop culture, Bruce will explain both how to create a Q-Learner (a popular type of reinforcement learner) as well as the philosophical significance of the existence of Q-Learning – and why it means Machine Learning has a brighter future than you have probably imagined.

About the presenter: