An empirical evaluation of active inference in multi-armed bandits

hello everyone welcome to actin flab live stream number 24.0 today is june 16th 2021 and we're going to be talking about this paper in empirical evaluation of active inference in multi-armed bandits i'm daniel and i'm here with blue hi awesome welcome to the active inference lab everyone we are a participatory online lab that is communicating learning and practicing applied active inference you can find us at the links here on this page this is recorded in an archived live stream so please provide us with feedback so that we can improve our work all backgrounds and perspectives are welcome here and we'll be following good video etiquette for live streams here at the short link you'll find all of the live streams and different series that we do in the communications unit of actin flab and today we're going to be contextualizing in the dot zero video for two upcoming discussions in the second half of june 2021 on the 22nd and the 29th when we have discussion 24.1 and 24.2 on this paper and hopefully with the authors joining today in actin livestream number 24.0 we're going to be trying to set some context and give an introduction to the following paper an empirical evaluation of active inference in multi-armed bandits by the authors listed here and the video is just an introduction to some of the ideas it's not a review or a final word it's kind of like a three-way intersection we have people who are maybe within the active inference community and looking to be exposed to some different areas like bayesian statistics or machine learning the second road is those who are coming from bayesian statistics or machine learning approaches and curious about active inference and then of course we hope that this will be exciting and interesting even if you're unfamiliar with active inference or machine learning we'll hopefully try to connect it to some broader questions in behavior and decision making more broadly we're going to walk through the aims and claims of the paper the abstract in the roadmap covering a few big questions and then we're going to go through all the figures and some of the key formalisms of the paper so that whether you read the paper or not you'll hopefully be in a good spot to ask questions and learn more and of course in the dot one and dot two in the coming weeks we'll be discussing this same paper so save and submit your questions and let us know if you'd like to participate or contribute in any way here we are on the paper itself which has a screenshot of the cover on this slide i'll uh read the aims and claims and then blue you can give a first thought on what you thought were kind of cool pieces about what they aimed for or claimed in this paper we provided an empirical comparison between active inference a bayesian information theoretic framework and two state-of-the-art machine learning algorithms bayesian upper confidence bound ucb and optimistic thompson sampling in stationary and non-stationary stochastic multi-armed bandits we introduced an approximate active inference algorithm for which our checks on the stationary bandit problem showed that its performance closely follows that of the exact version and hence we derived an active inference algorithm that is efficient and easily scalable to high dimensional problems so what was cool about that or what made you excited to make this dot zero this paper was really awesome it um it shows how active inference can be used to solve problems that are sometimes computationally intractable and really pretty difficult so i think that the authors did a great job of deriving this active inference algorithm and proving that it's useful agreed and they did an awesome job bringing it analytically like with equations in line or at least being juxtaposed to other approaches in machine learning rather than appealing to a qualitative body of theory which is also great this is definitely one where the claims are specific and exact and that's what we'll be following up on…