Four papers on Neural coding, Predictive processing, and Cognitive modeling.

Hello and welcome everyone. It is November 2nd, 2023, and Blue and I are here in ACTIM live stream 56.0, kicking off the 56 series on a set of four papers. Welcome to the Active Inference Institute. We are a participatory online institute that is communicating, learning, and practicing applied active inference. You can find us at links here. This is a recorded and an archived live stream, so please provide feedback so we can improve our work. All backgrounds and perspectives are welcome and will follow video etiquette for live streams. Head over to activeinference.org if you want to learn more about the institute, get involved in any activities or learning groups. Let's begin with a little bit of an introduction and warm up, say hello, and then maybe just one thing that we found interesting about these papers or that made us want to jump in for the 56 series. So I'm Daniel. I'm a researcher in California, and I was very interested to learn about what the relationship is between different neural inspired approaches to machine learning and active inference. Blue? Blue? Blue? I mean, same, this kind of always like the neurobiologically inspired machine learning is always kind of something that piques my interest and I'm still here, still excited about it. And in this like set of papers, I was actually like, particularly interested in like sparsity and how that is maybe something that's particularly biological or not, or just kind of learning and diving deeper into sparsity as like a topic of consideration. Cool. All right. Well, in this series and in this video, we're going to discuss four papers, and we'll get to why we have four soon. Four papers are the Neural Coding Framework for Learning Generative Models, 2022. Convolutional Neural Generative Coding Scaling Predictive Coding to Natural Images, 2022. Spiking Neural Predictive Coding for Continually Learning from Data Streams, 2023. And CogNGen, Building the Kernel for Hyperdimensional Predictive Processing Cognitive Architecture, 2022. And as there are four papers, we're going to use acronyms for each paper and also go through them in this order. And again, we'll return to why there's four and why this order. The four acronyms are NCF-22, CGN-C-22, SNPC-23, PPCA-22. So the video is going to be an introduction. We're going to cover the big questions and then basically jump immediately into the four papers. And this is just an introduction. Feel free to write comments or if you're around in time to join for these upcoming discussions that we're going to have with Author. So, Blue, what big questions are you excited about here? So I think I already talked about sparsity, but I would like to know what other types of tasks that are perhaps maybe non-visual could Neural Generative Coding be suited to? And then the concept of, in the last paper, we'll get to it, of a memory echo as a prediction. I think, is prediction just a memory echo? And then that leads into what is the difference between I predict, or I expect, or I hope, right? I predict that the Eagles might win the Super Bowl with maybe 50 to 60%. That's the probability, right? 50 to 60% probability that the Eagles win the Super Bowl, but I totally hope that the Eagles win the Super Bowl 100%. So where's the difference between hoping? And I can actually, I thought about it and it's like action, the role, you hope something happens, you hope you get into a good college, you predict that you'll go to college, you hope you get into a good college, you get good grades, do all the things, and then you can act based on that hope. But I actually can take no action to affect whether or not the Eagles win the Super Bowl. So action is not necessarily a key role there, but how is hope different from prediction and expectation? And how would you approach modeling something like that? It's just something that derailed me anyway. That's interesting. Yep. So my big questions first, just on the scope. How is it different now that we're in…