Some thoughts by Glen Berseth

I have spoken on many reasons for the RLC, however there is one that is more forward looking that is important to point out. That reason is that Deep RL is hard, decision making is hard, making environments is detailed. Tool creation, coding, understanding in this space is more involved. Not only does DeepRL have the challenges of supervised learning but it must also accomplish similar goals under distribution changes, with longer training times, and evidence is far more noisy. Addressing these practical challenges are just as important as foundational work. This is evidenced by the foundational questions that could not be answered until recently because of infrastructure advances. Like the stone and flint, the combination of these two materials start fires that advance research.

Then there is a question left unanswered. How to best make progress on these challenges There are two primary jobs I have as a professor and researcher, one is to perform the best research I can and address the toughest questions that will have the most impact for the betterment of society. The other is to create spaces for passionate people to do the same and pursue their own interests, and in this case a space for the people in the RL community to focus on these more complex problems and work together to find solutions to accelerate the community. There is no doubt that there are people with the skills or the passion to learn the skills, and I have seen many times the success of putting the right people together, in the same room, and having them work the problem. I am very excited to see all of you at the RLC, and look forward to the barriers we burn down together.