Demystifying Deep Reinforcement Learning @NVIDIA GPU Tech Conference — Silicon Valley

The Nividia GTC is a yearly AI conference at Silicon Valley, Europe, China and Japan. I usually attend the conference. This year I am also teaching a lab on Reinforcement Learning — super excited !

It is in 2 parts — the agenda is somewhat linear — Part 1 ends with DQN and Part 2 is on Policy Gradients.

It is a lot easier to listen to a lecture or a video, but Reinforcement Learning has lots of nuances. So I have marked the equations and the attendees should write the formulas — not one but three times, preferably with a Kuru Toga mechanical pencil !

Then only one will get a feel for the components of an equation, where the groupings, parenthesis, the limits of sigma’s and so forth.

There is an excellent book on Teaching Tech Together and wanted to remind myself the important pragmas …

Programming experience is expected, but no prior knowledge of Reinforcement Learning.

Lots of people to thank for.

We start with a good framing of Reinforcement learning

Discuss challenges and compare with other ML methods

Learn how to skate Frozen Lakes in Siberia, without falling into holes

We have quizzes

We dive deep as well as look broader with Taxonomy discussions

Because we have only 4 hrs there are things to explore outside the class !

About 80% of the materials is in the Deep Learning Land

We unroll the math as needed

We discus AlphaGo, A3C and DDPG in addition to MonteCarlo, SARSA, Q-Learning, DQN and REINFORCE

And, … end with a question …

So bring your sharp Kuru Toga mechanical pencils (er … actually the engine sharpens the lead as you push it!) & even sharper minds … to GTC 2019 Silicon Valley Part 1 & Part 2 …

The materials are in my github. Would appreciate updates, insights, comments and corrections as pull requests …