A Practical Introduction to ReinforcementLearning.jl
▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔
I am glad to welcome Jun Tian @findmyway as our speaker, who will be joining remotely from Beijing. Because of 6 hours difference in timezone we start earlier this time.
Jun Tian is the main developer and maintainer of ReinforcementLearning.jl. He will give an introduction to the ecosystem and can also answer all your deep question.
The level will be a mixture of beginner & intermediate.
┉┉┉┉┉┉┉┉┉┉┉┉┉┉┉┉AGENDA┉┉┉┉┉┉┉┉┉┉┉┉┉┉┉┉ 16:00 - 17:30: Jun Tian: A Practical Introduction to ReinforcementLearning.jl
Reinforcement learning has led to many breakthroughs in recent years.
In this talk, Jun Tian will give a short introduction on how to use
ReinforcementLearning.jl to solve some interesting problems. After
this talk, users will be able to apply modern deep reinforcement
learning algorithms to the problems they are interested in.
17:30 - 18:00: Time for further discussion
┉┉┉┉┉┉┉┉┉┉┉┉┉┉┉┉┉┉┉┉┉┉┉┉┉┉┉┉┉┉┉┉┉┉┉┉┉┉┉
The event is going to be online. The link will be made available 10 min before the actual start. Please join 5 min ahead, the session is going to start on-time, Monday 02. May 2022 16:00 CEST.
Thank you for this meetup. As always great organization. Very interesting topic and presentation. @schlichtanders Would you please pass the following question to the presenter: “Would you please consider organizing something like an online minisymposium on RL in a foreseeable future: (ReinforcementLearning.jl, AlphaZero.jl, POMDPs.jl …)”
I am very sorry, I was not aware of this Julia Discourse feature to find a person by a real name. Thank you @schlichtanders for rising this topic and many thanks to @findmyway for such interesting presentation. (This is written with a smile.) I would like to add that in my opinion minisymposium on reinforcement learning would speed up the development of activities of this particular area of Julia Community, which I believe is at the forefront and is one of the most interesting fields of scientific developments currently taking place, with many real world applications.
“Would you please consider organizing something like an online minisymposium on RL in a foreseeable future: (ReinforcementLearning.jl, AlphaZero.jl, POMDPs.jl …)”
I talked with @jonathan-laurent last summer on some potential directions to integrate RL.jl and AZ.jl. I think our conclusion at that time was that except for the environments part we only have very few components to share with each other. But we reached an agreement that we do need some better tools for distributed computing in RL specific fields. And that’s the main motivation of OpenTelemetry.jl.
Although these packages are all RL specific and have some overlap, I think we are all exploring very different directions. For me, I’m trying to split out some common components into independent packages that may be useful for others, like CircularArrayBuffer.jl, or Trajectories.jl. So that others can reuse them to implement very different algorithms.
I would like to add that in my opinion minisymposium on reinforcement learning would speed up the development of activities of this particular area of Julia Community, which I believe is at the forefront and is one of the most interesting fields of scientific developments currently taking place, with many real world applications.
I totally agree. Like I said in the talk, I think I will be more well-prepared for such mini-symposium after this summer once I finish the next release of RL.jl (it’s been more than half a year since the last release). And hopefully, I can share some industry-level experiences on RL at that time.
I understand. Although my experience so far is mostly related to AlphaZero.jl, I will look forward with anticipation for the next release of RL.jl. I will also take a closer look into OpenTelemetry.jl and its implementation prepared by the author of AlphaZero.jl, especially that 1.8 might be close to RC and AFAIK should bring significant improvments related to BLAS. To refer to your presentation, I do hope that the OpenTelemetry could be one of the ways leading to unification of the resources. I also do fully agree with you that there is not that great overlap, on the other hand it seems that there is a significant potential for cooperation. I’d like to say thank you once again and I do hope that the topic of reinforcement learning shows up on the agenda of JuliaUserGroupMunich one more time (hopefully more then once) in the future.
Monday 02. May 2022 
16:00 CEST.
17:30 - 18:00: Time for further discussion