Lilian Weber

Looking forward to this!

Are you coming to @rldmdublin2025.bsky.social #RLDM2025?
Join our workshop on Gamification in Computational and Cognitive Neuroscience with an amazing speaker lineup! 🎮👾🧠💫

w/ @ondrejzika.bsky.social @ingrdmrtn.bsky.social

The interoceptive origin of reinforcement learning Rewards play a crucial role in sculpting all motivated behavior. Traditionally, research on reinforcement learning has centered on how rewards guide l…

For the past couple of years I met weekly with 3 amazing younger scientists @lilweb.bsky.social @debyee.bsky.social @rikepetzschner.bsky.social to share ideas and teach each other. No agenda. It was often the favorite part of my week. Then we started writing. www.sciencedirect.com/science/arti...

Re-posting is appreciated: We have a fully funded PhD position in CMC lab @cmc-lab.bsky.social (at @tudresden_de). You can use forms.gle/qiAv5NZ871kv... to send your application and find more information. Deadline is April 30. Find more about CMC lab: cmclab.org and email me if you have questions.

Two days left for contributed talks submissions 🤖💫

sites.google.com/view/game-on...

so, so excited to share our new theory which endeavours to answer why depression and poor metabolic health go hand-in-hand: "An interoceptive model of energy allostasis linking metabolic and mental health" 💪🧠🔋 @saramehrhof.bsky.social
@hugofleming.bsky.social

osf.io/preprints/ps...

📣 Exciting announcement for the summer term:

Starting in May, the @ccnberlin.bsky.social seminar series by Felix Blankenburg & Timo Schmidt (FU Berlin) is co-hosted by @david-s-stolz.bsky.social David Stolz (Universtiät Lübeck) and myself.

The topic of all talks is: "CONTROL"

🚨💫🤖 RLDM Workshop announcement:

"Game on: Leveraging Gamification in Cognitive and Computational Neuroscience"

Contributed talks submissions now OPEN: shorturl.at/0p0Jd

Looking forward to seeing you in Dublin!
w/ @luiantaverra.bsky.social & @ingrdmrtn.bsky.social

Interesting paper from Millridge et al: Reward Bases: A simple mechanism for adaptive acquisition of multiple reward types

Do DA neurons encode gradient of reward dimensions, i.e some encode food predicting cues, others fluid etc.?
journals.plos.org/ploscompbiol...
#compneurosky #neuroskyence

YouTube video by cutting EEG CuttingEEGX - Evolving Methods

My 15' talk "the modern EEG regression toolkit: rERPs, TRFs, mixed models, and more" is now on youtube youtu.be/vnlbVDarRUc?...

thanks @cuttingeeg.bsky.social for an awesome #EEG conference!

Yes!!! The review I've been looking for! Rapid-acting antidepressants like ketamine, electroconvulsive therapy, and noninvasive stimulation (TMS) are thought to work by enhancing plasticity, but how, exactly? Grateful to @alexkwan.bsky.social and team for summarizing the evidence in this review.

Check out this👇 amazing new python tool for Bayesian modelling, lead-developed by @nicolaslegrand.bsky.social - it includes (among many other things) all elements of the generalised HGF (arxiv.org/abs/2305.10937) for you to use and apply. Can't wait to see what the community will do with it! 🚀

Eli Sennesh, Hao Wu, Tommaso Salvatori
Divide-and-Conquer Predictive Coding: a structured Bayesian inference algorithm
https://arxiv.org/abs/2408.05834

We have a new preprint out, with Teresa Schuhmann, Alex Sack and Catherine Tallon-Baudry.

I think our findings give some interesting leads on the various mechanisms underlying visceral influences on the motor system and how we conceptualize interoception as a whole.
#neuroskyence

Altered food liking in depression is driven by macronutrient composition of the depicted food. We analyzed wanting and liking ratings of patients with major depressive disorder (MDD) and matched healthy control participants. Patients with MDD reported lower food wanting, whereas food liking was primarily associated with symptoms of anhedonia. We then evaluated whether the macronutrient composition of the food accounted for differences in ratings. This led to a substantial improvement in models and showed that patients with MDD show specific deficits for food rich in fat and protein.

Can we use macronutrients to understand altered food reward in depression? Our new preprint shows that patients with MDD report lower wanting & liking for foods rich in fat or protein (vs. carbs). #neuroscience 🩺

W/ @corcerebrum.bsky.social @neuromadlab.bsky.social
www.medrxiv.org/content/10.1...

This has been a long haul – the main work was done as part of my PhD at the #TNU @ethzurich.bsky.social , with Klaas and Chris. Years later, @nicolaslegrand.bsky.social, Peter and Anna came onboard and created shiny toolboxes from our initial codebase. First preprint May 2023. Glad to finally share!

We also provide some examples for the kinds of models you can now build within the HGF framework – from global volatility beliefs, multimodal inference, multisensory cue combination over time, etc etc – check it out! arxiv.org/abs/2305.10937

There’s plenty more in the preprint. For example, we find that the message passing implied by value hierarchies in the HGF is mostly equivalent to what has been proposed for PC, with small, but interesting differences.

GitHub - ilabcode/pyhgf: PyHGF: A neural network library for predictive coding PyHGF: A neural network library for predictive coding - ilabcode/pyhgf

Even better, we implemented the gHGF in two(!) shiny and openly available toolboxes for you:
1. ✨ In Python (led by @nicolaslegrand.bsky.social): github.com/ilabcode/pyhgf
2. ✨ In Julia (led by Peter Thestrup Wade): github.com/ilabcode/Hie...

This modular architecture of the generalised HGF means that it’s now super simple to build your own models under the HGF – you can literally add or remove nodes in your belief network and don’t need to worry about deriving any new equations! 🚀

Here, we generalise the HGF to encompass *both* value and volatility belief hierarchies, all under the same framework, with one-step belief updates driven by precision-weighted prediction errors, and a modular belief network architecture.

In contrast, PC typically focusses on hierarchies in which higher level beliefs concern the #value (mean) of lower levels. This is useful for understanding how beliefs about lower-level features depend on context (higher-level beliefs) (e.g., the classical Rao & Ballard, 1999).

In current HGF models, higher level beliefs encode the #volatility of lower levels. This is motivated by the observation that learning should be influenced by environmental uncertainty: if the world is currently changing (volatile), the agent needs to learn faster. 🏃‍♀️

Predictive coding and the HGF are both hierarchical Bayesian models, but they tend to assume different kinds of belief hierarchies: volatility coupling (HGF) versus value coupling (PC).

🤔🧐
What does this even mean??

The generalized Hierarchical Gaussian Filter Hierarchical Bayesian models of perception and learning feature prominently in contemporary cognitive neuroscience where, for example, they inform computational concepts of mental disorders. This...

Long overdue thread on: ✨The generalised hierarchical Gaussian filter (gHGF)✨

Ever wanted to build your own HGF without deriving/implementing new equations? Ever wondered how volatility beliefs in the HGF relate to predictive coding?

Check this out:
arxiv.org/abs/2305.10937