Wouter Nuijten

Looking for prospective students & collaborators to work towards *An Agent Foundation Model*! 👨‍💻👩‍💻 Keywords: Active model discovery, active learning & model based planning. More details: lancelotdacosta.com/research (bottom of page). If you’re interested, get in touch! 💡

I'm in Copenhagen for @EurIPSConf this week, showing some work at the Epistemic Intelligence in ML workshop. If you're in town and want to link up, let me know! ​#EurIPS #MachineLearning #AI #Copenhagen

🏆 Best Technical Paper at @iwai_ws! Recasting EFE planning as Variational Inference → sample-efficient Active Inference agents that actually run. Built in RxInfer. Won by @wouterwln. Paper: arxiv.org/abs/2508.02197 Code: github.com/biaslab/EFEasV… #IWAI2025 #RxInfer #Bayesian

Recently at @lazydynamics, we needed a fine-grained scheduler, coupled to the machine clock, to build our multi-scale agents. To streamline development, I built `Gears.jl`, available now at github.com/lazydynamics/G…. We chose to make it public, so let me know what you think!

What's even nicer: because our method injects priors locally, everything still works within @ReactiveBayes ' RxInfer.jl using message passing. Special thanks to my colleagues at @LazyDynamics for making this possible!

This is what we ask ourselves in our latest paper (arxiv.org/abs/2508.02197), where we solve Minigrid environments with Active Inference. (Spoiler, it works!) Code available at github.com/biaslab/EFEasV… Done together with @mlmykola , @ThijsvdLaar and @bertdv0

In Active Inference, a lot of time is spent on computing Expected Free Energy. What if we could tweak the generative model such that EFE can be minimised with traditional variational inference methods?

Agreed with @fchollet on FEP (t.ly/s4gMV), but FEP is more than a pretty good idea, and there are more benefits to realizing an agent as an active inference (AIF) process beyond active data selection. I will mention a few below:

New paper alert! Active Inference (AIF) agents score policy candidates using a cost function called the Expected Free Energy (EFE), which has many desirable features, such as a parameter-free balance between information-seeking and goal-driven behavior. (1/4)

Turns out .ml doesn’t just stand for machine learning - it also means maybe lost :/ But hey, every project needs its "we lost our domain" story. Our just happened on Friday. Find us at rxinfer.com now. This time for real (probably).

Catch me at @NeurIPSConf in Vancouver, BC this week! I will be at the conference all week, so feel free to hit me up for a drink (or five)! Saturday, I will present a poster at the Bayesian Decision Making workshop.

Exciting work from @ReactiveBayes developers (@mlmykola, @wouterwln, @bvdmitri, İsmail Şenöz) presented at the NeurIPS Bayesian Workshop! Riemannian Black Box VI (RBBVI) tackles non-differentiable, costly models. 📄 openreview.net/pdf?id=QBbc0L5… 💻 github.com/biaslab/Gradie…

This work is together with @mlmykola , @bvdmitri , Ismail Senöz and @bertdv0

Absolutely thrilled to announce that our work “Riemannian Black Box Variational Inference” has been accepted at the NeurIPS Workshop on Bayesian Decision-Making and Uncertainty! See you in December in Vancouver ;) openreview.net/pdf?id=QBbc0L5…

Also, this week I am at @iwai_ws in Oxford to talk about Active Inference, so hit me up for a Free Energy Chat ;)

A couple of weeks ago, I talked at @JuliaConOrg about one of my packages. The video is available here: youtu.be/7DFmyzgNeGQ?si…

🚀 Calling all Bayesian enthusiasts! @ReactiveBayes needs YOU to shape the future of reactive, efficient & scalable inference. Join us in building Bayesian-driven autonomous agents with RxInfer.jl. Fork, contribute, discuss at rxinfer.ml #BayesianInference #Julia

Catch me at @JuliaConOrg in my hometown of Eindhoven this week! I will be presenting twice on Friday and strolling around the Philips Stadium tomorrow and Thursday !

Today, I was a guest in a workshop hosted by @InferenceActive on RxEnvironments.jl, demoing the basics of the package and answering lots of interesting questions. Recording is available at youtu.be/ygmAhA18mkQ?fe…. Thanks Daniel & co for having me!

@AlbertPdsnk @sirbayes @dmdohan @bvdmitri For example, we know the order in which variables are created, and we know that people tend to write their generative models in the same order as they'd want to see it drawn (left to right, top to bottom). We could use that heuristic as a starting point

@AlbertPdsnk @sirbayes @dmdohan @bvdmitri Honestly, I think we can make a serious attempt towards this. We have a lot of implicit knowledge about the graph structure without needing additional user specification. Would be interesting to see what a naive solution would look like.

This is my semi-periodic request to VLM gurus: when will I be able to upload an image such as the one below and have {ChatGPT, Gemini, Whatever} give me back tikz code that reconstructs the figure? I would pay >$20/month for that! (@dmdohan knows :)