TalkRL: The Reinforcement Learning Podcast

RL podcast is all reinforcement learning, all the time. Featuring brilliant guests, both researched and applied. Join the conversation on Twitter at talk r l podcast. I'm your host, Robin Chauhan. I'm very glad to be here with professor Satinder Singh, one of the founders of RLDM Conference.

We're here at RLDM twenty twenty five, which is at Trinity College in Dublin, Ireland. Professor Singh, can you tell us the origin story of RLDM?

Yes. My PhD adviser, Andy Bartow, was retiring from the University of Massachusetts, Amherst, and we were had organized a retirement event for him. And a number of us were there, Rich Sutton, Yael Niv, a bunch of Andy's Barto students. And after that event, we were taking a taxi back from Amherst to the airport in Hartford, which is the closest airport.

And we got to talking about how great Andy Barto has been for the field, how he's connected psychology, neuroscience, and and machine learning. And Yael was in the car. I was in the car. I think Rich Sutton was in the car. And we we we talked about how there was no meeting that really brought these communities together in an organized way.

And, you know, just wishful thinking about how great it would be to have such a meeting. And Yael mentioned that Princeton has a program by which you can propose new conferences, and there was an easy way, lightweight way of proposing this, and she was very confident that we could get a decent amount of money to fund such a conference. And so that's how it started. We we we the fact that there was an easy way to get money without making putting a lot of effort in then allowed us to write a write a quick document and bring a few people together as a founding team. I may not get every name right.

I think it's Rich Sutton, Peter Dayan, Yael Niv, and me. I apologize to those of whom I'm not remembering. And then then we just started thinking about what the meeting should look like. We wanted it to be a meeting. Not a conference.

You know, computer science is very driven by papers and archival papers and references and citations and so on. And the neuroscience and psychology side really writes journal papers, and they don't really value conferences quite that much. So we wanted to set up a kind of a thing which isn't like the rest of the machine learning conferences, where people would submit abstracts, and there would be lightweight reviewing, but and really kind of a very curated set of talks that both from what we call the dry side, which is computer science and machine learning, and the wet side, which is neuroscience and psychology. We also wanted to get roboticists and control theorists on the table as well. Anyway, that's how the conference started.

And can you say a bit more about how RLDM has evolved over the years?

Yeah. Maybe first, I'll begin by adding a little bit of backdrop. I think we again, as I said in the previous answer to the previous question, we deliberately decided to make RLDM a meeting and not a conference. Over the years, a lot of times, the question of whether we should have an RL conference has come up, and at least the senior people have always felt that it's not a good idea to have a reinforcement learning conference because it's better to be engaged with the larger community, you know, the AI community, the robotics community, the machine learning community, the neuroscience community, the psychology community, and that we should present our work at those conferences, like NeurIPS and ICML and and AAAI and and so on. So we deliberately didn't make this a conference.

We made it a meeting. But, you know, over time, a new generation of reinforcement learning researchers are asking the same question, have asked the same question. Should we should we have a reinforcement learning conference? And we had a first one in Amherst last year. And so now we have a reinforcement learning conference.

And one one question for the RLDM meeting is, you know, what does it mean to now have a two two RL focused meetings, the reinforcement learning conference and a reinforcement learning and decision making meeting. And so that's posing a question and a challenge, you know, for for RLDM. And, you know, there are different views on this. But, certainly, my own view is that we should double down on the fact that this is a meeting where we have neuroscientists and psychologists and computer scientists, and we are here to have conversations and and and and talk to each other. And so this will still remain the home, I I hope, for this really integrally multidisciplinary set of people.

But I think reinforcement learning conference will will become a conference is a conference where we'll publish sort of archival reinforcement learning papers. And but but there is this question now. We have two RL meetings, and can the field sustain two two RL meetings?

Thank you, professor Satinder Singh.