Geometric Methods for Sampling, Optimisation, Inference and Adaptive Agents

Hello and welcome. This is ActInf livestream number 52.0 and it is February 28th, 2023. Welcome to the Active Inference Institute. We're a participatory online institute that is communicating, learning, and practicing applied active inference. You can find us at the links here on this slide. This is a recorded and an archived livestream, so please provide feedback so we can improve our work. All backgrounds and perspectives are welcome and we'll be following video etiquette for livestreams. Head over to activeinference.org if you want to learn more and participate in learning groups and projects at the institute, including these livestreams. Well, we're here today to learn and discuss the paper, Geometric Methods for Sampling Optimization Inference and Adaptive Agents by Alessandra Barp, Lancelot DaCosta, Guilherme Franca, Carl Friston, Mark Ghirilami, Michael Jordan, and Gregorios Pavliotis. This video, like all .0 videos are, is an introduction for some of the ideas. It is not a review or a final word. And as we've joked before, it's more than anything, a call for help. So if you're curious about these topics, if you're knowledgeable about these topics, we would really look forward to you getting involved in the upcoming 52.1 and 52.2 discussions, as well as in an ongoing basis to help us understand some of the technical details. This is going to get technical at times, and certainly beyond the technicalities I understand, though I am looking forward to presenting them. And it will be great to have those who have backgrounds of all different types to come together and talk about this awesome work. Also, big thanks to Ali and Kandon for the assistance, technical and moral, during the preparations here. In 52.0, we're going to bring up some aims and claims, read the abstract, look at the roadmap, and talk about the keywords. And then we will walk through the paper with an emphasis on the figures, formalisms, and key points. In the coming weeks, we're going to be discussing this paper with one or more authors. So, as usual, get in touch if you want to participate, or if it's after the fact, you can still get involved. We can start with an introduction or a warm-up. I'm Daniel. I'm a researcher in California. And I'm tempted to say, just totally honestly, I'm happy to get this one over with, but that sounds a little bit different than I might intend it to be. I'm really excited to dive into this work, which is going to be approaching active inference from an angle that we haven't necessarily highlighted on these streams. So, I think it's going to be a fascinating discussion. It's going to run the gamut, span the gap, however you choose to see it, between technical sophistication and intuition, which is a great place to be. So, I've been really excited and motivated to prepare, and I'm looking forward to the .0 we're doing right now, and to the upcoming discussions. So, let's jump in with the big questions that might motivate one to read this paper, this kind of paper, even if they were not familiar with the authors or topics. And these are just a few ways to write it up, of course not the only ways. So, first question. How can we effectively and efficiently navigate information geometric or information theoretic landscapes? And how can we tackle that question from an analytical, which is to say, equation based, as well as a computational, which is to say, real implementation based perspective? Often, it's really fun, intuitive, natural to talk about information theory, even for those who haven't taken the technical prerequisites. And this work may help us navigate to a space where we're able to think with good intuitions about information geometry, information theory, and also make sure that those intuitions are going to be caught by our technical tools. Second, how can we optimize inference in complex models, including cases where we are doing inference on action as a parameter, also known as active inference? Optimization and…