Practical RL

Agent Smith: Never send a human to do a machine's job. (c)

Machine Learning methods went through a steep ascension over the last decades. Given enough labeled data, one can teach an algorithm to find objects on images, comprehend and even generate text and voice, translate natural language or retrieve information from the internet at near- or superhuman levels. The only downside is, not every problem can be framed as learning X -> y transition that approximates some reference labels.

If you find yourself learning to, say, ride a bicycle, play some new game, navigate in an urban surroundings, design landing pages or, heck, build reinforcement learning agents in this very course, normally you don’t just memorize textbooks of examples of optimal muscular contractions for every possible situation. The common idea of these problems is that they can be solved by trial and error: trying out ideas and sticking to those that hurt less.

One more common thing is that these problems can be, to a various extent, solved automatically. Yup, the creative search for solution is exactly what we’re going to train machines to do throughout this course.

The main focus of the MOOC are the practical aspects of training such “machines”, called Reinforcement Learning (RL) algorithms, for a life-size problems. Things on the menu: foundations of RL, practical algorithms, engineering “hacks”, case studies, fresh & crunchy articles. The schedule features a variety of stuff from games and robotics through finance to chatbots.

One more thing to know is that this course will have a tight connection with deep learning methods. There’s no strict requirement to have neural networks experience as we’ll be giving a crash course on them (using Theano+Lasagne), but knowing how to cook neural networks will definitely come in handy time to time.