#GSO2025

Closing #GSO2025, Qianyang Chen explores how to interpret the famous Ising model as a perception-action loop.

After coffee break, Mikhail Prokopenko discusses the biological arrow of time, a recent paper he has published:
arxiv.org/abs/2409.12029

#GSO2025

At #GSO2025, @amahury.bsky.social presents our work with @hirokisayama.bsky.social on The Origin and Evolution of Information Handling.
Preprint at arxiv.org/abs/2404.04374

Now Jan Dohmen discusses how we can infer context in reinforcement learning generalization.

#GSO2025

Opening the last #GSO2025 day, David Wolpert talks about the relationship between statistical physics, computation and information. He gave us a lot of valuable insights on how to apply stochastic thermodynamics to the phenomenon of communication.

Before poster session Claire Glanois talks about Self-Organized Language Agents (SOLA), an innovative approach to model political polarization. Her results show that SOLA exhibit self-organizing behaviors very similar to the those obtained in segregation dynamics.

#GSO2025

Now Andreas C. Schneider discusses the limits of information theory when quantifying complexity. Particularly, he proposes Partial Information Decomposition as a way to encode information redundant between feedforward and context. Very interesting triadic approach!

#GSO2025

After coffee break, Hyppolite Charvin proposes symmetry as a principle to formalize sensimotor perception. His approach looks for symmetry detection using information theory. I'd like to know what would be the implications of Noether's theorems from this perspective!

#GSO2025

Now Nihat Ay discusses how we can improve the evidence lower bound when dealing with training generative models. It remains to be explored how these results could be applied to the brain predictive machine hypothesis.

#GSO2025

After lunch Bernhard Schölkopf talks about the importance of structural causal models to represent mechanisms in digital twins (DT). The way he displayed representations is identical to Rosen's modelling relation.

#GSO2025

Before lunch, Vladimir Jácimović discusses how to translate machine learning approaches to the hyperbolic domain. This hyperbolic ML opens the door to exploring in which fields these mathematical tools could be useful.

#GSO2025

Now @vvh.bsky.social discusses how embodied agents are useful when exploring models of developmental cognition. Particularly, her research focuses on refining measures of agency and self applied to tool-use tasks.

#GSO2025

Now Pierre Schumacher discusses the emergence of natural and robust bipedal walking by learning to control up to ninety muscles from biologically plausible objectives! His results show that embodiment and bio-inspired reward lead to natural motion.

#GSO2025

Kicking off #GSO2025 's second session, Martin Riedmill talks about current efforts to achieve general AI, emphasizing the importance of reinforcement learning to build embodied reasoning agents capable of solving multiple tasks. 🤖

Currently Michael Harré gave a very thought-provoking talk about how to outline a path toward collective intelligence using AI and advances in Theory of Mind. Definetely we need to use human cognition as a potential direction towards general AI. #GSO2025

To finish today's session Max Mittenbühler discussed the behavioral effects of automatic response tendencies and meta-control emergence as a product of maximizing evidence. #GSO2025

After poster session we have Martin V. Butz discussing how active contextualization (AC) may overcome the frame problem. He showed interdisciplinary evidence proving this, and then he showed how context alleviates cognitive efforts and proposed AC to develop general AI.

#GSO2025

Just discussed the most recent version of my paper with @hirokisayama.bsky.social and @cgershen.bsky.social at the #GSO2025 poster session. Very happy to find people interested in this project! I'll be giving an oral talk about it this Friday, so stay tuned. 😼

Now @cansusancaktar.bsky.social talks about curious exploration in reinforcement learning via semantically sensible exploration. This framework allows us to explore meaningful behavior in vastly different environments. The next step is to reach open-endedness via parallel learning. #GSO2025

Before lunch Anna Levina discusses the role of criricality in foraging agents. Her results tell us that populations of agents evolve to a slightly subcritical state. Additionally, the distance to criricality depends on the complexity of the task.

#GSO2025

Now we have Bulcsú Sándor talking about a toy model for a self-organized sensorimotor loop in a robot. His results point to the emergence of self-organized locomotion by switching between dynamical attractors, producing proprioceptive feedback with the environment.

#GSO2025

Our first keynote speaker is Sebastian Risi, who talked about multiple examples of agents able to change their morphology to adapt their behavior. His next step is to explore the organism-environment relationship, which could give us more insights into self-organization. 🦠

#GSO2025

No #complexitycat blog this week as I'm attending #GSO2025 in Tübingen! I will share with you my experience through this channel. Looking forward to presenting my progress in the origin and evolution of information handling + enjoying the lovely keynote speakers we will have. 😸