Quantitative Promise Theory: Intentionality and Inference in Autonomous Agents - Mark Burgess Theoretical Neurobiology Group 2026-07-14 | A brief overview of Promise Theory and how it may relate to Free Energy/Active Inference. Promise Theory grew out of a desire to understand autonomous cooperation, and has existed for 20 years. It is widely used in technology, and has been applied to sociology, biology, and physical problems.Mark Burgess is a British theoretical physicist, turned multi-disciplinary Computer Scientist. He was Professor of Networks and Systems at Oslo University College, now an independent researcher and consultant at ChiTek-i, based in Oslo, Norway. (see http://markburgess.org)
Radical Surprise - Leonidas Pantelides Theoretical Neurobiology Group 2026-09-11 | Title: Radical Surprise: Sufficiency Conditions for Framework Reconfiguration in Active Inference—The Constitutive Role of Time.This talk examines the conditions under which models remain coherent in the face of “radical surprise”—cases in which the frame within which prediction operates becomes insufficient. Drawing on diverse examples, it introduces the notion of transcontextuality as the capacity to reconfigure relevance across local and non-local temporal frames when the system comes under internal pressure. I will show that the Active Inference Model accommodates framework reconfiguration and show how this apparent violation of the Gödel Incompleteness theorems may be explainable along the lines that, implicitly, the Model operates as a temporally open system, grounded in non-local conditionality rather than a closed formal structure. Leonidas Pantelides is an active diplomat presently writing a philosophical book on the problematics of time.
Active Inference as theTest-Time Scaling Law for Physical AI Agents – Omar Hashash Theoretical Neurobiology Group 2026-07-17 | In this talk, we will uncover how active inference can serve as a test-time scaling law for physical AI agents (e.g., autonomous vehicles, humanoid robots, etc.). In particular, this scaling law enables physical AI agents to reason in the unforeseen scenarios that appear at test time to generalize, analogous to large language models (e.g., GPT-5) that reason to improve their response to harder prompts. Up to this end, we merge reinforcement learning (RL) and active inference to showcase that policies naturally scale with inference under a unified formulation of narrow task and general survival rewards. Effectively, this alleviates the stationary assumptions that have constrained RL at test time, to transform it into a continual learning scheme in the real world.Bio:Omar Hashash is a postdoctoral research associate at Virginia Tech, USA, where he earned his PhD in 2025, with research interests encompassing artificial intelligence (AI), world models, digital twins, wireless communications, and active inference (website).Christo Thomas is an assistant professor at Worcester Polytechnic Institute, USA, where his research interests include causal inference, Bayesian optimization, mathematical foundations of AI, and semantic communications (website)Chair: Robert Chis-Ciure
Quantifying Phase Transitions in Human Development and Psychopathology - Dimitris Tsomokos Theoretical Neurobiology Group 2026-06-30 | ABSTRACTHow can experience that accumulates gradually give rise to behaviour that abruptly reorganises itself? Active inference can help answer this from first principles. In a minimal two-state POMDP with Dirichlet learning, when we couple parameter learning to action selection (i.e., the chosen strategy decides which evidence is sampled, and that evidence shapes the next choice), the model reduces to a self-consistency equation under a mean-field closure. That equation is formally identical to the mean-field Ising model, so a developmental stage shift becomes a bifurcation condition and the cusp catastrophe long postulated in developmental psychology is derived rather than assumed. We verify this in a balance-scale model that recovers Siegler's classic findings, then extend the same machinery to adolescent depression. In the depression model, we find that onset can be abrupt even under gradual adversity, and recovery is governed by the motivational field rather than cognitive control; this ordering is empirically tested in two large birth cohorts from the UK and US.PRESENTERDr Dimitris I. Tsomokos currently works on child and adolescent mental health and developmental psychology at the UCL Institute of Education, with a previous research background in theoretical physics (quantum information theory and condensed matter).Google Scholar: scholar.google.com/citations?user=e7khjO4AAAAJ&hl=en&oi=aoCHAIRRobert Chis-CiureGoogle Scholar: scholar.google.com/citations?user=7V9C7skAAAAJ&hl=en&oi=ao
Information as Maximum-Calibre Deviations in Kinetic Ising Models - Alexander Kearney Theoretical Neurobiology Group 2026-06-28 | In the study of learning, consciousness, and cognition, many theories overlap substantially yet remain distinct in their methods, terminologies, and research protocols. Deep specialization within each has yielded immense value, and presents a compelling opportunity for each field to be strengthened by insights from others. Kearney presents the case that studying systems as they deviate from (internally 'predicted') maximum-calibre (MaxCal) path ensembles offers a bridge. In particular, Integrated Information Theory's (3.0) mathematics can be derived by applying successive constrained MaxCal procedures. Bayesian Mechanics' dualism with CMEP and the proposed G-theory are well-established in the literature. Further, recent findings that Fluctuation Dissipation Theorem (FDT) violations empirically mark consciousness (Berjaga-Buisan et al., 2025) stand amidst a tradition of theoretical work suggesting critical, unstable, non-equilibrium states characterise the conscious brain. Kearney explores these ideas in Kinetic Ising models (Ishihara and Shimazaki, 2025) by applying the MaxCal procedures derived from IIT 3.0 to construct a minimal definition of information (or its precursor) over long paths. He then applies Large Deviations Theory to construct a variational free energy functional which, he argues, is created by the system's internal variables and minimised during states of (dynamic) equilibrium. Sensory data, he suggests, validates its use by coupling internal self-regulation with exterior inference. Information as first order prediction error arises, alongside equivalence between VFE minimization and a supervised learning problem. This motivates the use of stochastic gradient descent to minimize the objective function, which yields spiking patterns and predictive coding style update rules. In the case of a single (recurrent or non-recurrent) "neuron", FDT-style equations can be derived from the dynamics induced, and offer a natural timescale over which inference should occur. While the nuances of each theory remain distinct, he suggests that Kinetic Ising models offer a meeting ground against which IIT, FEP, and non-equilibrium theories of cognition can be compared, assessed, and perhaps unified, in precise, mathematical terms.Alexander Kearney completed a master's thesis in Integrated Information Theory at the Mathematical Institute, Oxford University, and has since collaborated with industry researchers on AI uncertainty quantification. Most recently, he's working with Florida Atlantic University on empirical measures of consciousness through Neuromatch's Sentience Scholars program.
Interactive Inference: A Simplified Active Inference Account of Interaction - Roel Vertegaal Theoretical Neurobiology Group 2026-06-21 | In this presentation, I explore simplifications to Active Inference that make its central intuitions more accessible to designers, HCI researchers, and other non-mathematical users. The core proposal is that many simple interactive behaviours can be approximated by treating variational free energy locally as a second-order Taylor approximation around equilibrium. Under this approximation, the dynamically relevant part of surprise takes the form of a quadratic potential, half S/N squared, where S denotes error and N denotes the noise scale. The ratio S/N gives normalized prediction error: the number of discriminable choices. As in energy-based physics, we ignore the Z partition function (negative log model evidence). What matters for behaviour is the gradient, relaxation dynamics, and the threshold at which residual surprise becomes acceptable. This allows perceptual and action-oriented updating to be modeled as descent on the normalized surprise potential, with a stopping criterion defined by a minimum acceptable KL divergence expressed in noise units. The same quadratic form is equivalent to the potential energy of a Hookean spring and to the negative log likelihood of a Gaussian, up to a constant. For two Gaussians of equal variance, its readout corresponds to the KL divergence between prior and posterior. Bayesian updating can therefore be interpreted as relaxation of a charged surprise potential toward equilibrium, yielding logarithmic time expressible in terms of Shannon's theorem. I will show how this model can predict classic laws in human-computer interaction. In Fitts’ Law, hand movement time becomes a logarithmic reduction in surprise toward a target edge. In Hick’s Law, logarithmic response time arises by loading a surprise potential with the number of unresolved options and relaxing it until one choice reaches a threshold of one. More complex forms of this decision model can yield a competitive relaxation process approximating softmin-like policy selection. Initial results suggests that variational free energy and expected free energy may be less sharply separated in behaviour than in formal exposition if the surprise potential involves simulated behaviours during policy selection. Prof. dr. Roel Vertegaal is Professor of Human-Computer Interaction and Director of the Human Media Lab at Radboud University in Nijmegen, The Netherlands. Elected to the ACM SIGCHI Academy in 2022 for his contributions to HCI, he was trained as an interaction designer, computer scientist, cognitive scientist, and electronic musician. He is a pioneer of mobile interaction technologies including the foldable phone, attention-aware user interface, holographic video conferencing, and interactive drone swarms, with over 150 refereed publications, numerous patents, and broad popular press coverage.
Dr. Edward Owens and Dr. Pedro Sanz, 8th of June, A Novel Perspective Through the Lens of the FEP. Theoretical Neurobiology Group 2026-06-19 | Abstract:When viewed through the lens of the Free Energy Principle (FEP), sleep bruxism—a sleep‑related movement phenomenon characterized by rhythmic masticatory muscle activity—emerges as an active inference policy through which orofacial motor activity minimizes high‑precision interoceptive prediction errors during sleep, modulated by sleep stage and anxiety. By defining Markov blankets at their multiple scales—from brainstem and cortical circuits to masticatory muscles and airway biomechanics—we describe how perception–action cycles minimize prediction error about bodily states during sleep. Sympathetic bursts, cortical micro‑arousals, respiratory disturbances, comorbid conditions, and (dental) occlusal factors are reconceptualized as modulators of prediction error driving orofacial motor policies within an integrated autonomic–arousal–motor complex. We introduce the construct of homeostatic latency, describing the temporally extended window over which rhythmic masticatory muscle activity reflects the efficiency of restoring quasi‑stable equilibrium after arousal‑related perturbations. Finally, we propose a model within a domain‑specific knowledge graph architecture that encodes these mechanisms, providing a computationally tractable and clinically actionable framework that generates testable predictions for polysomnographic, electromyographic, and clinical indices and supports individualized, mechanism‑informed assessment and management.Speaker Bio: Dr Edward Owens, B.A., B.Dent Sc., M.S.D., FACP, Diplomate of the American Board of Prosthodontics International Certificant American Board of Dental Sleep Medicine Beacon Consultants Clinic, Sandyford, Dublin www.beacondentalsleep.ie; Dr Pedro Mayoral Sanz, DDS, MS, PhD, Director of the Dental Sleep Medicine Master Program, Universidad Católica de Murcia UCAM, Madrid, Spain.TNB website:https://www.activeinference.institute/tnb
A machine that learns from its own experience - Rahul Chouhan Theoretical Neurobiology Group 2026-06-10 | This talk reports an eleven-month pre-registered empirical program on B3 — unsupervised regime-change detection — treating it as the load-bearing prerequisite for any artificial agent that genuinely learns from its own experience under an Active Inference framework. After five mechanistically distinct failures of cluster-failure detection in three-state toy environments, I moved the identical detector into a rich semantic environment with a small (7B-class) language model as the world-model substrate, with frozen pre-registered acceptance criteria on a controlled 40-episode protocol. Qwen2.5-7B-Instruct and Phi-3.5-mini-instruct both achieve 77.5% detection at zero false alarms — class confirmed at n=2 across two vendors and two model sizes — with Mistral-7B serving as a mechanism-diagnosed negative control ("confident prior leakage," cheap remedy empirically excluded); the entire result is reproducible on free Colab T4 GPU. The proposed thesis—scale the world, not the model—is that the binding constraint on unsupervised continual learning is environmental richness rather than parameter count or detector design, and the open question I bring to the group is what the right free-energy-framework formalization of "environmental richness" would be and whether the crossing generalizes beyond curated semantic environments into genuinely open-ended embodied settings. What else can help "a machine that can learn from its own experience"?
Cortical circuits for predictive processing - Georg Keller Theoretical Neurobiology Group 2026-06-03 | I’ll present evidence for the different circuit elements necessary to implement predictive processing in cortex (partially following this pubmed.ncbi.nlm.nih.gov/30359606), and will finish on why I think the model is incorrect (partially discussed here pubmed.ncbi.nlm.nih.gov/38424472/).apredictiveprocessinglab.org
Hierarchical Active Inference for Abstract Task Learning and Spatial Navigation - Toon Van de Maele Theoretical Neurobiology Group 2026-06-02 | Schema learning refers to the ability of humans and other animals to acquire abstract task representations that capture structural regularities shared across experiences. This capacity enables generalisation to new environments, adaptation to changing rules, and planning in complex situations. In the paper, the authors propose that interactions between the medial prefrontal cortex and the hippocampus can be modelled as hierarchical active inference (HAI), in which abstract task representations interact with lower-level spatial inference and navigation processes. In this presentation, he will illustrate this idea through two examples. First, he will present a model in which HAI separates task structure from navigation, showing how their interaction enables an agent to solve structured navigation tasks with repeating action sequences, and how disrupting communication between hierarchical levels impairs performance. Second, he will show how this framework can be extended to demonstrate that abstract task representations form schemas that generalise across contexts. In this model, a probabilistic mapping (grounding likelihood) links abstract goals to spatial locations, allowing schemas to be flexibly deployed in new environments while remaining grounded in spatial inference. Together, these examples illustrate how hierarchical active inference supports flexible planning and generalisation across tasks and contexts.June 1, 2026
Nir Asch, 26. May 2026, From Dysconnection to Intervention: Schizophrenia Theoretical Neurobiology Group 2026-05-26 | From Dysconnection to Intervention: A Pallidal Bottleneck for Precision Routing in Schizophrenia?Abstract: This talk asks whether schizophrenia-related dysconnection may have a controllable circuit bottleneck — the external globus pallidus (GPe) — for precision routing across cortico–basal ganglia loops. I will connect active-inference accounts of aberrant synaptic gain to exploration–exploitation behavior, focusing on reduced directed exploration and increased random exploration as behavioral readouts of precision dysregulation. I will then review evidence suggesting that the GPe is a plausible precision-routing hub and present non-human primate data showing that NMDA-R antagonism disrupts exploration balance, while low-frequency GPe macro-stimulation restores it. The talk concludes by asking whether GPe-targeted neuromodulation could provide a translational route from dysconnection theory to intervention in schizophrenia. Speaker Bio: Nir Asch, MD-PhD, I am a psychiatry resident at Rambam Medical Center and a computational neuroscientist at the Hebrew University of Jerusalem, studying cortico–basal ganglia mechanisms of cognitive flexibility, psychosis-relevant inference, and neuromodulation. TNB Info: The TNB Group has been fostering interdisciplinary research and collaboration for decades. Our mission is to advance the understanding and application of active inference, a theoretical framework developed by Prof. Karl Friston. This is achieved through regular online meetings featuring presentations and discussions, which may include theoretical frameworks, empirical data and its analysis, simulations, and mathematical development. We welcome contributions and perspectives from diverse fields, including neuroscience, mathematics, machine learning, psychology, philosophy, medicine, and biology.Translational Neurobiology Group: https://www.activeinference.institute/tnb
Excitatory-inhibitory resonance in cognition stabilizes synaptic traces in memory - Don Tucker Theoretical Neurobiology Group 2026-05-20 | We outline a dual-control theory of motivational control of the sleep process throughout ontogenesis in which excitatory (approach) and inhibitory (avoidance) waking concerns persist into sleep as striatal-limbic resonances—electrophysiological oscillations with affective charge from subcortical controls that carry forward the processing demands of unresolved experience. These two opponent motive controls function as operators of active affordance in sleep, with NREM inhibitory control differentiating the constraints implied by recent experience and REM excitatory control generating predictive options for adaptive action. Don M. Tucker is a neuropsychologist at Brain Electrophysiology Laboratory Company, and a professor emeritus at the University of Oregon.
The Self as a Dynamic Model: Psychiatry, Prediction, and the Fragility of Mind - Erik Smedler Theoretical Neurobiology Group 2026-05-14 | In this talk, I explore the self as a biologically grounded yet dynamically constructed phenomenon. Drawing on clinical psychiatry, developmental neurobiology, evolution, and predictive processing, I examine how agency, meaning, and stability emerge — and how they sometimes fail.This lecture is part of my ongoing work integrating philosophy, neuroscience, and clinical experience in a forthcoming book on the brain and the mind.Erik Smedler is a physician-scientist from Gothenburg, Sweden working as a psychiatrist and leading a research group focusing on molecular psychiatry. With background in physics, he has a general interest of dynamical systems and theoretical neuroscience, although the current research is focused on the use of stem cell-based models of the nervous system. Read more here: www.smedlerlab.org
Individually-Centric Cognition via Hierarchical Active Inference - Thomas Loker, 21.04.2026 Theoretical Neurobiology Group 2026-04-24 | We present ICFA, a cognitive agent architecture built on the Free Energy Principle where each layer of the hierarchy runs as an independent active inference agent at its own native scale, communicating through inter-blanket outcomes rather than coarse-grained data. We introduce a Confidence Corridor that extends the ELBO with a Confidence Upper Bound Estimate (CUBE), providing a bounded operating envelope for self-limiting autonomous behavior, and provide directional evidence that hierarchical compositionality emerges from quality-gated layer formation without explicit renormalization operators. The ICFA Cognitive Lab — a live simulation environment — demonstrates agents learning to navigate grid, sparse, and maze environments from scratch with no pre-training, forming composites that self-assemble and earn promotion to recursive FractalCores based on proven predictive accuracy. We conclude with a live demonstration and open questions on the relationship between ICFA's structural coarsification and RGM renormalization, and on extending these principles to hidden-state inference for behavioral biometric authentication.Thomas W. Loker is the Co-Founder and CTO of Individual Centricity Corporation, where he and Co-Founder Juval Löwy have spent over twenty years developing individually-centric architectures, publishing on the Free Energy Principle, precedence-based causality, and consciousness, and now applying Active Inference to on-device behavioral biometric identity validation through IC-Corp's 1TrueU fractalCore platform.The TNB Group has been fostering interdisciplinary research and collaboration for decades. Our mission is to advance the understanding and application of active inference, a theoretical framework developed by Prof. Karl Friston. This is achieved through regular online meetings featuring presentations and discussions, which may include theoretical frameworks, empirical data and its analysis, simulations, and mathematical development. We welcome contributions and perspectives from diverse fields, including neuroscience, mathematics, machine learning, psychology, philosophy, medicine, and biology.https://www.activeinference.institute/tnb
How do temporal and content-based predictions shape visual processing? - Gil Kfir Theoretical Neurobiology Group 2026-04-22 | This study aimed to investigate how content-based (“what”) and temporal (“when”) predictions interact to shape visual perception within a predictive processing framework. Using MEG data from 30 participants viewing sequences of rotating Gabor patches, the study applied an inverted encoding model (IEM) to decode orientation signals and compare neural responses across conditions varying in temporal and orientation congruency. Contrary to the hypotheses, no significant interaction or simple effects of temporal prediction were found, suggesting limited evidence for combined predictive influences under the current design. The findings highlight potential methodological constraints, including limited statistical power, and raise questions about how temporal precision and motor-related processes might be more effectively examined in future research. Gilad Kfir is a neuroscientist and organizational consultant in practice. In his research work he is currently affiliated with Universite Paris Cite where he studies the effects of psychedelic experiences on therapeutic outcomes throughout insights and belief updating processes. Monday, March 2, 2026
The Geometry of Discovery: Locating Scientific Structure Beyond LLMs and Knowledge Graphs-S. Rahman Theoretical Neurobiology Group 2026-04-15 | Description: We present a discovery engine that extracts governing mathematical structure initiated from scientific corpora using topological and geometric methods. Seeded with 2.5M+ arXiv papers, the system produces a living manifold distinct from both the stochastic interpolation of LLMs and the rigid ontologies of traditional knowledge graphs, identifying 130 emergent mathematical fields and over 36,000 structural gaps across domains. We demonstrate how the manifold dynamically reorganizes around user context, and as a case study, examine where Active Inference is structurally located on the manifold relative to its parent fields in variational calculus, stochastic dynamics, and Bayesian inference. We also present applications in multi-agent simulation (extending standard POMDPs with mean-field stochastic coupling) and non-equilibrium macro-finance.Presenters:Vladimir Baulin, PhD is Chief Scientist at the Synthetix Institute and architect of the Discovery Engine protocol. Distinguished Researcher at Universitat Rovira i Virgili with 150+ publications and over 8,000 citations in soft matter physics, biophysics, and complex systems. Board member of the Active Inference Institute. His recent work includes a framework for AI-driven synthesis of scientific knowledge landscapes and a lead contribution to the Intelligent Soft Matter roadmap (2025).Shagor Rahman is Head of Product at the Synthetix Institute, where he leads the design of the Atlas research platform. His work spans product design and management, with publications in Frontiers on transcendent model selection and cognitive simulation methods.Also joining from the Synthetix team: Andrew Pashea (Chief Agentic Architect), Austin Cook (Head of Engineering), and Jonathan Minchin (COO).
Active Attentional Inference for Multi-Attribute Choice Under Uncertainty - David Hyland Theoretical Neurobiology Group 2026-04-14 | ...
Pranjal Balar, 07.04.2026 - Precision-Weighted Arbitration Between Interoception And Exteroception Theoretical Neurobiology Group 2026-04-09 | This talk presents a predictive processing model in which interoceptive and exteroceptive signals are treated as parallel hierarchical inference streams whose influence on perception and emotion is determined by precision-weighted arbitration. The framework formalises how relative confidence assigned to bodily versus environmental prediction errors shapes affective experience and regulatory behaviour. Using analytical formulation and simulations, the model reproduces stable, flexible, and maladaptive regimes corresponding to affective phenotypes. The work generates testable predictions for behavioural, physiological, and neuroimaging data, situating affective dysregulation as a problem of precision control rather than sensory abnormality.Pranjal Balar is a second-year BSc Neuroscience student at UCL and research fellow at the Swasth Foundation and is currently working on computational model of interoception-exteroception balance, precision weighting, and affective regulation.The TNB Group has been fostering interdisciplinary research and collaboration for decades. Our mission is to advance the understanding and application of active inference, a theoretical framework developed by Prof. Karl Friston. This is achieved through regular online meetings featuring presentations and discussions, which may include theoretical frameworks, empirical data and its analysis, simulations, and mathematical development. We welcome contributions and perspectives from diverse fields, including neuroscience, mathematics, machine learning, psychology, philosophy, medicine, and biology.https://www.activeinference.institute/tnb
Tangential Action Spaces and Active Inference - Marcel Blattner Theoretical Neurobiology Group 2026-04-07 | ...
Predictive Activity in Language Processing - Luigi Grisoni Theoretical Neurobiology Group 2026-03-31 | This talk will examine predictive activity in language processing and its theoretical implications for current models of language processing. I will first introduce recent EEG evidence showing early interactions between semantic, pragmatic, and phonological factors across language modalities, consistent with distributed and modality-independent representations rather than asymmetric or partially segregated accounts. I will then present a series of studies on perception, language comprehension, and production demonstrating anticipatory predictive signals that vary as a function of stimulus predictability and recruit also preferential sensory and motor brain systems depending on the expected critical event. Taken together, these findings support distributed models of language processing and highlight the role of predictive mechanisms in shaping language processing across modalities.Luigi Grisoni is a researcher in cognitive neuroscience and psycholinguistics. For over 13 years, he worked at Freie Universität Berlin with Professor Friedemann Pulvermüller, investigating the neural bases of language, meaning representation, and predictive processing. Since September 2025, he has been based at the Istituto Italiano di Tecnologia (IIT) in Ferrara, where he works with Professor Luciano Fadiga and Professor Alessandro D’Ausilio.
Observer Theory - Sam Senchal Theoretical Neurobiology Group 2026-03-31 | A detailed breakdown of the computational Observer Model. We will discuss the theory, the priors, the properties of an Observer, the formalism (the Ruliad), its links to FEP and the concept of surprise, the measures vs FEP (probabilistic) and the applications of Observer Theory to Evolution, Telos (Formal Causation) and EthicsSam A Senchal is an independent researcher, loosely associated with the Wolfram Institute focused on computational philosophy, conciousness, metaphysics and theology. You can find more work on Observer Theory at observertheory.substack.com and github/sassenchal.
Justice for all Majid Beni Theoretical Neurobiology Group 2026-02-28 | This presentation posits a model-based reinterpretation of Rawls’s theory of justice, arguing that the principles of justice emerge as an objective property of self-organising systems maintaining a non-equilibrium steady state (NESS). By framing concepts like reflective equilibrium as a dynamic NESS process of internal entropy regulation, the model directly addresses critiques regarding the lack of objective grounding in Rawlsian theory. We demonstrate that enduring societies, by implementing mechanisms analogous to Rawls’s principles of justice, effectively minimise variational free energy—a principle ubiquitous in biological and cognitive systems—thereby ensuring their functional stability and adaptive capacity across multiple scales, from the neuronal to the global. This synthesis anchors the political philosophy of justice within the established mechanisms of theoretical biology and statistical physics.Majid D. Beni is a philosopher of science and serves as an Associate Professor of Philosophy at Middle East Technical University, Ankara. Link: sites.google.com/site/majiddavoodybeni/?pli=1Chair: Robert Chis-Ciure, scholar.google.com/citations?hl=en&user=7V9C7skAAAAJ&view_op=list_works&sortby=pubdate
Let there be information - Hector Manrique Theoretical Neurobiology Group 2026-02-12 | Abstract: Brainstrom about how the capacity to rein in information across vast temporal scales—thousands to millions of years—can help us trace the evolution of cognition. I propose that the ability to bound surprisal serves as a proxy for cognitive evolution, functioning as a proximate mechanism.Bio: I am Professor of Developmental Psychology at University of Zaragoza (Spain), I seek to answer Evo-Devo questions by applying Active inference and the FEP.
Neural Policy Composition from Free Energy Minimization - Francesca Rossi Theoretical Neurobiology Group 2026-02-02 | The ability to flexibly compose previously acquired skills to execute intelligent behaviors is a hallmark of natural intelligence. This capability is commonly attributed to gating mechanisms that regulate how multiple policies, or primitives, are weighted and composed. In this talk, a principled and theoretically-grounded model in which gating rules emerge from the minimization of free energy is proposed. From this, a continuous-time dynamical system that provably converges to the optimal solution is derived, together with a neural implementation as a soft-competitive recurrent neural circuit. The model is evaluated on collective behavior and human decision-making benchmarks, reproducing key behavioral signatures and providing insights into the data.January 26, 2026
A Theory of the Mechanics of Information - Christopher Hazard Theoretical Neurobiology Group 2026-01-13 | Abstract:Active inference and the free energy principle provide top-down accounts of learning, framing the learner as a system that minimizes prediction error. Here we introduce a complementary bottom-up approach built on information theory that performs inference directly from data, requiring no hyperparameters, and attains at or near state-of-the-art performance across nearly all machine learning tasks. We then dissect the information‑theoretic trade‑off between exploration and exploitation and embed it in a formal model of creativity. Finally, we map neural‑network architectures onto this framework and outline how it could be integrated with active inference and free‑energy theory to advance a unified “physics of information.”Bio:Dr. Chris Hazard is co-founder and CTO of Howso, with a cross-domain background in computation, AI, information theory, game theory, and nonlinear dynamics.
Excitatory-Inhibitory Resonance Establishes Cortical Memory - Don Tucker Theoretical Neurobiology Group 2025-12-19 | We propose that experience and memory emerge from structured resonance between excitatory (E) and inhibitory (I) population waves in the cerebral cortex, constrained by cortical geometry and organized by subcortical controls. This mechanism of E–I resonance links deep (layer 5-6) excitatory waves, generating limbifugal predictions, with superficial (2-3) inhibitory networks, mediating error-correction on the basis of sensory input in predictive coding formulations of active inference. When the low-frequency (theta) excitatory waves resonate with in-phase (n:m coupled) higher frequency (gamma) inhibitory waves, the resonance reflects shared information and creates transient standing‑wave “write windows” that persist long enough to induce early LTP in the local synaptic network.December 9, 2025
Studying the emergence of agency in cellular automata within the FEP framework - Michele Vannucci Theoretical Neurobiology Group 2025-12-03 | The objective of this research is to investigate the minimal conditions and properties under which generalized cellular automata (GCA) can effectively display adaptive behavior within their environment, while reinforcing the links between machine learning, free energy, chaos theory and self-organization. The proposed setting features a pre-defined perception-action interface, which acts as a Markov blanket between the environment and a GCA with a set of mixing and structural properties. The latter are tuned for the GCA to exhibit adaptive behavior consistent with the Bayesian and energetic interpretation of the free energy principle. Moreover, the emergent behaviour is constrained to satisfy classical RL criteria, for it to (efficiently) converge to an optimal policy, and information theoretic criteria: for the internal states to encode a compressed and meaningful representation of the hidden causes. The research will have a mathematical focus to derive some proofs in this direction, complemented by numerical analysis where needed November 4, 2025
A Biologically Plausible Model of Individual Differences in Decision-Making - Dayoung Yoon Theoretical Neurobiology Group 2025-12-03 | This research introduces and validates a new decision-making model based on the Active Inference Framework with the objective of better explaining individual behavioral differences and their neural underpinnings in psychiatric conditions like suicidal risk. To validate the model, we used a probabilistic learning task, analyzing fMRI data from healthy volunteers and behavioral data from a clinical group with Major Depressive Disorder (MDD) and varying levels of suicidal risk. Our results demonstrate that the AIF model outperforms conventional reinforcement learning models in explaining both behavioral choices and correlated brain network activity. Most notably, model parameters related to heightened sensitivity to negative outcomes successfully distinguished high suicidal risk individuals from non-suicidal individuals within the MDD group, highlighting the model's potential clinical utility.December 1, 2025
A Prosomeric Model Account of the Evolutionary Basis of Active Inference - Phan Luu Theoretical Neurobiology Group 2025-12-03 | The prosomeric model of brain organization has implications for understanding the evolutionary basis of active inference. We present how an understanding of brain organization from this prosomeric perspective provides insights into the brain systems that underlie exploitative and explorative drives for information seeking. These systems are rooted in mesencephalic and diencephalic “motor” centers, and telencephalic evolution extended these systems into modes of control seen both in the organization of the basal ganglia and cortex.Speaker:Phan Luu is a neuropsychologist and electrophysiologist (bel.company).December 02, 2025Chair: Robert Chis-Ciure
Two Opportunities and Two Questions for the Active Inference community - Dan Wilkinson Theoretical Neurobiology Group 2025-12-01 | This talk will introduce a real-world immediate opportunity for development and deployment of Active Inference: adaptive management and optimisation of complex heterogeneous workloads running on large scale classical HPC infrastructure. It will then re-introduce the concept of probabilistic bits (p-bits), an emerging new hardware concept that may offer huge benefits for Bayesian and Active Inference in relation to the task above. Finally, two open questions will be posed to the audience: (1) How the task described above can be properly formulated as an Active Inference problem and how can humans be incorporated in the loop? (2) What quantity of p-bits at some given accuracy and reliability might be needed to make this Active Inference solution for this task a reality in the medium term and critically - what might the programming paradigm look like?November 24, 2025
A Generative Geometry of Collective Intelligence - Nicolás Hinrichs Theoretical Neurobiology Group 2025-11-25 | How do interacting brains become a “shared mind”? In this talk, I will present a formal active-inference model of two agents engaged in real-time interaction, where each continuously updates its model of self and other to minimise surprise. Within this framework, affective states emerge from each agent’s inference about its own identity coherence (prediction errors) and the shifting boundary between self and other. To test these ideas empirically, we introduce geometric hyperscanning, a novel approach using network curvature to track moment-by-moment changes in inter-brain connectivity. This method reveals phase transitions in shared neural dynamics (e.g. moments of rupture and re-attunement in mutual understanding). By bridging a formal model with multi-brain data, the work offers new insights into how collective intelligence arises from interpersonal exchange, and it points toward a multi-scale approach to shared cognition across natural and artificial agents.Speaker:Dr. Nicolás Hinrichs is a researcher at the Okinawa Institute of Science and Technology (Japan) and the Max Planck Institute for Human Cognitive and Brain Sciences (Germany), where he combines active-inference theory, multi-brain neuroimaging (hyperscanning), and information geometry to study how interacting agents develop shared (collective) intelligence. (Homepage: nicolashinrichs.de)November 25, 2025Chair: Robert Chis-Ciure
From Charles Darwin’s “Root Brain” to Nikola Tesla’s “6G World Brain” - Nika Hosseini & Martin Maier Theoretical Neurobiology Group 2025-11-19 | ...
Generalised Electrodynamical Model of Mitochondrial Oxidative Phosphorylation - Anthony Finbow Theoretical Neurobiology Group 2025-11-11 | Explorations to characterize the function of the Mitochondrion within the Eukaryotic cell have largely focused on its central role in generating ATP; “the energy currency of life”. Mitochondria are here reconceptualized as utilising sophisticated bio-electromagnetic oscillators, which dynamically stabilize their function and form synchronised oscillatory networks. This proposal outlines a novel framework where mitochondrial cristae act as resonant cavities, electron transport chains (ETCs) function as phase-locked loop (PLL) reference oscillators, and ATP synthases serve as (driven) voltage-controlled oscillators (VCOs) and demodulators. We propose that fundamental biophysical forces – coherent electron flow, self/mutual inductance, gyroscopic dynamics, and Lorentz forces – govern energy transduction, enable active sensing of local environment and spatial attitude, mediate rapid "electronic" switching, and orchestrate precise metabolic control through coupled oscillations with the Krebs cycle.Anthony Finbow is an independent researcher and entrepreneur and Faculty Member at The Guy Foundation for Quantum Biology.November 11, 2025
Minimizing the regularization term of VFE = maximizing variational empowerment - Yusuke Hayashi Theoretical Neurobiology Group 2025-11-10 | The key ideas that will be covered in the presentation are reflected in this blogpost: lesswrong.com/posts/AndYxHFXMgkGXTAff/universal-ai-maximizes-variational-empowerment-new-insightsNovember 3, 2025
Universal Surprise-Minimizing Agents - Michele Vanucci Theoretical Neurobiology Group 2025-11-05 | Abstract: We explore how an agent with unlimited computational resources, namely AIXI, behaves when it is given the goal of minimizing surprise over incoming observations. This intrinsic reward is motivated by the FEP, and implemented in only a few works within the reinforcement learning literature. By removing the problem of computational constraints, we can investigate whether the sole goal of surprise minimization leads to interesting or trivial behavior. As we’ll see, this heavily depends on the Kolmogorov complexity of the environment and how easy it is to predict with the Solomonoff mixture. In this setting, we find that the optimal behavior would be for the agent to find a niche of the environment which is easily predictable, without altering the prior weight on simpler environments significantly. Nonetheless, we argue that the surprise-minimizing agent would still exhibit exploratory behavior in chaotic environments, where it is harder or impossible to find a static niche.Speaker: Michele Vanucci. This study constitutes the final thesis for my Master of Science in Artificial Intelligence at Vrije Universiteit Amsterdam, it is supervised by Peter Bloem. My personal website: michelevannucci.me.November 4, 2025TNB Chair: Robert Chis-Ciure
Active Inference and Proprioceptive Deafference - Andrew King Theoretical Neurobiology Group 2025-10-28 | How does Active Inference help us to interpret symptoms in rare cases of loss of proprioception after viral infection? I will present an interpretation of the movement symptoms seen in people who have lost proprioceptive sense, as they have been reported in published case studies. I will derive some hypotheses about possible anatomical substrates of limited changes in symptoms following the sensory loss and pose some questions about how we integrate different sensory modalities for motor control as viewed under Active Inference.Dr Andrew King is a lecturer in physiotherapy at the University of Liverpool. He has clinical experience in neurological rehabilitation, vestibular rehabilitation, and an interest in gait analysis and motor control.October 28, 2025TNB Chair: Robert Chis-Ciure
Cognitive Resource Theory: A Gauge-Theoretic and Thermodynamic Model of Cognition - Kyle Cahill Theoretical Neurobiology Group 2025-10-28 | This presentation will introduce Cognitive Resource Theory (CRT), a physical theory that models cognition as a thermodynamically constrained system governed by real-time dynamic energy allocation and reallocation. CRT was developed to formalize the physics underlying a candidate hypothesis for how behaviorally relevant neuroplastic adaptations can emerge from prolonged cognitive engagement in sensory-rich environments, such as action video games. I will present empirical evidence from multimodal neuroimaging, including structural, functional, and directed connectivity, that inspired CRT, followed by an overview of its formal construction. I will conclude with a discussion of potential interfaces with the Free Energy Principle and outline future directions for testing and theoretical generalization.Kyle Cahill, PhD, is a neurophysicist specializing in brain connectivity and network dynamics. https://www.physics.gsu.edu/cahillOctober 27, 2025
Active Inference under Feedback Limitations: Systems Theory and Cancer Treatment - Emre Koksal Theoretical Neurobiology Group 2025-10-21 | In this presentation, we discuss the foundations and applications of active inference under conditions of limited feedback. We begin by deriving the free-energy principle in the context of dynamical systems and provide an information-geometric interpretation of free-energy minimization on the probability simplex. The connections between free-energy minimization, fundamental limits of feedback control systems, and directed information from observations to latent states are highlighted. Building on this foundation, we present an application to oncology, where cancer treatment is modeled as a constrained decision process. Here, the patient is viewed as a dynamical system, and treatment policies are designed to minimize tumor progression under strict measurement and clinical burden constraints.October 20, 2025TNB Chair: Robert Chis-Ciure
Active control of internal representations in spiking NNs & robots - Alejandro Rodriguez Theoretical Neurobiology Group 2025-10-08 | In this talk, I will discuss our recent work on active inference in two of the areas of expertise in our group. In the first one, we show a simple neural circuit capable of self-evidencing using continuous representations and the consequences of different architectural motifs and particular partitions. In the second one, we present a different kind of internal model used for human-robot or robot-robot interaction. We introduce the concept of informational coupling as means of reproducing dyadic dynamics observed in humans and animals. We conclude discussing our results in the light of their relations with some areas of physics and ecological psychology, presenting also some ongoing ideas and future work.July 28, 2025
What is an artifact? Active inference and technology - Luca Possati Theoretical Neurobiology Group 2025-10-08 | The goal of this contribution is to explore how Active Inference and the Free Energy Principle (FEP) can provide conceptual tools to deepen our understanding of the human-machine relationship, socio-technical systems, and, above all, design. My argument unfolds in two interconnected parts. The first part claims that Active Inference and the FEP offer an elegant, parsimonious, and—most importantly—physics-based theoretical framework for justifying the extension of agency to artifacts. This is a crucial issue in both technology studies and design. My argument is as follows: In Active Inference, agency is essentially self-evidencing (Hohwy 2016, Friston 2013)—an emergent property of any system equipped with a Markov blanket. The very existence of a Markov blanket (i.e., a specific partition of states that ensures conditional independence between internal and external states, given the blanket states) entails agency as self-evidencing. This is far from trivial, since a Markov blanket isn’t just a label that can be arbitrarily applied to any system (Beck & Ramstead 2025, Thestrup Waade et al. 2025, Hesp et al. 2019). Therefore, for systems with a Markov blanket, being a “thing,” acting, and exchanging information in a particular manner are all intrinsically linked. Moreover, this form of agency should be seen as existing on a spectrum: the complexity and temporal depth of self-evidencing depends on the structure of the Markov blanket and the generative models. This gives us a highly flexible and physics-grounded notion of agency—one that can be extended to artifacts, as long as they possess a genuine Markov blanket. Philosophically, this opens up a rich dialogue with Bruno Latour’s Actor-Network Theory and with new materialism (Bennett 20120, Barad 2007, DeLanda 2016). The second part of my argument focuses on the notion of design. If we accept the premises laid out in the first part, then design can be understood as the act of creating and developing new forms of agency—that is, new Markov blankets and generative models. But what does it mean to create a Markov blanket?September 30, 2025
Language, Prediction, and the Origins of Understanding - Patrick Krauss & Achim Schilling Theoretical Neurobiology Group 2025-09-30 | Our talk explores how the human brain predicts syntactic structure during natural language comprehension, based on simultaneous recordings from magnetoencephalography (MEG) and electroencephalography (EEG) as participants listened to continuous speech. Our results show that different word classes, such as nouns and verbs, elicit distinct neural responses, and that anticipatory brain activity indicates sensitivity to upcoming syntactic elements. Additionally, we observe potential parallels between biological and artificial language processing, as words with higher predictability scores from large language models are associated with reduced N400 amplitudes. Questions: 1) We observe anticipatory brain activity before nouns appear in continuous speech. From the perspective of active inference, could this reflect not just passive prediction, but also an active process of preparing or shaping expected input? 2) In our data, the N400 amplitude is smaller for highly predictable words, consistent with reduced semantic surprise. Would you interpret the N400 as a direct reflection of a prediction error signal in semantic processing, or as a downstream process related to integration after prediction fails? 3) In our study, we found that nouns elicit stronger pre-onset neural activity and are more easily predicted by large language models, possibly reflecting syntactic priors in both systems. However, unlike the brain, LLMs operate without sensory grounding or action. From the perspective of active inference, how critical is embodiment or sensory-motor grounding for generating syntactic predictions, and do you see a path toward integrating this into future LLM architectures?29th September
TMS Assisted Psychotherapy - Toward an Interdisciplinary, Personalized Framework - Grant Brenner Theoretical Neurobiology Group 2025-09-24 | Transcranial Magnetic Stimulation (TMS) is an FDA-approved, noninvasive therapy for Major Depressive Disorder (MDD) and Obsessive Compulsive Disorder (OCD) in the United States. For MDD, TMS typically targets the left dorsolateral prefrontal cortex using high-frequency pulses across multiple sessions to improve mood and reduce symptoms. TMS works by stimulating nerve cells in brain regions involved in mood regulation, activating frontoparietal networks and promoting neuroplasticity via a range of mechanisms. While TMS is effective for symptom relief, research is ongoing to determine how best to combine it with psychotherapy and related approaches to support more lasting, developmental change. Neuronavigation, which uses brain imaging to personalize TMS targeting, offers potential for tailoring treatments to individual patients and expanding TMS applications to more complex or less standardized clinical presentations. This presentation will overview the current state of TMS for psychiatric applications, discuss the relevant literature on combining TMS with psychotherapy, propose a model for TMS-Assisted Psychotherapy (TAP), and briefly touch upon neuronavigation and emerging approaches using AI to develop protocols and guide targeting, seeking input on development of both TAP and AI assisted targeting using emerging machine learning paradigms, and briefly touch upon additional modalities such as Transcranial Focused Ultrasound Stimulation (tFUS) and Temporal Interference (TI), which allow for non-invasive modulation of deeper brain structures . Overview of TMS-Assisted Psychotherapy: psychologytoday.com/us/blog/experimentations/202505/tms-assisted-psychotherapy-moving-toward-a-paradigm-shift23rd September 2025
Numadelic Virtual Reality: A paradigm for design beyond day-to-day phenomenology - David Glowacki Theoretical Neurobiology Group 2025-09-23 | Virtual Reality (VR) within psychology and neuroscience research is typically used to construct computational renderings of ‘what things look like’ – i.e., which maintain fidelity to day-to-day perception. In this talk, I will describe how virtual reality (VR) can be used to design experiences beyond day-to-day phenomenology. Specifically, I will describe our work to implement a ’numadelic’ aesthetic within VR, which allows people to visualize their own bodies and the bodies of others as light energy, with diffuse spatial boundaries. Despite their minimal phenomenology, numadelic representations produce a strong sense of presence & immersion, with associated emotional and somatic sensations. Combining numadelic representations with cloud-mounted VR technology enables completely new classes of embodied experience like “coalescence” – e.g., where the bodily representations of multiple people can completely overlap, reliably producing poignant psycho-somatic effects. From the perspective of predictive coding, numadelic representations appear to challenge top-down priors for how people imagine their bodies in relation to others. More broadly, VR approaches like these offer an experimental platform for better understanding various aspects of predictive coding.Presenter: David GlowackiDavid Glowacki, PhD, MA (www.glow-wacky.com), is the founder of the ‘Intangible Realities Laboratory’, a research group at the Research Center for Intelligent Technologies in Santiago de Compostela, Spain.Literature:1. Glowacki, D.R., VR models of death and psychedelics: an aesthetic paradigm for design beyond day-to-dayphenomenology, doi.org/10.3389/frvir.2023.12869502. Glowacki, D.R., Williams, R.R., Wonnacott, M.D. et al. Group VR experiences can produce ego attenuationand connectedness comparable to psychedelics. (2022). doi.org/10.1038/s41598-022-12637-zTNB Chair: Robert Chis-Ciure
Reframing lived experience of persistent dizziness: insights from active inference - Marc Broberg Theoretical Neurobiology Group 2025-09-23 | We present an active inference account of persistent dizziness that is informed by the patient’s lived experience, as interpreted through enactivist and ecological psychology models. Under this theoretical framework, we explore holistic, principled, and person-centered approaches to therapeutic intervention. 22nd September 2025
Emotion as Rhythmic Inference: Toward a Unified Framework for Affective Resonance - Juyoung Kim Theoretical Neurobiology Group 2025-09-19 | This talk introduces a conceptual framework treating emotion not as a static state but as a dynamic, rhythm-like phenomenon—a “fluid of affective inference.” Drawing inspiration from both quantum theory and complex system dynamics, we propose a model where emotional resonance emerges from interactions between internal predictive flows and external synchrony. While detailed structural and mathematical formulations are currently under patent review, the presentation will focus on the philosophical foundations and potential implications for dynamic emotional memory, affective disruption modeling, and the design of relational AI systems. Our aim is to open an interdisciplinary discussion bridging affective neuroscience, philosophy, and computational cognition.9th September 2025
The Hidden Architecture of Intelligence: One Pattern. Everywhere - Max Kelly Theoretical Neurobiology Group 2025-09-17 | What is the universal pattern behind intelligence: from the first living cells to human societies to the machine minds of the future? Starting with the simple question of how anything persists in a chaotic universe, this talk takes a clear narrative journey from basic physics to the shared architecture that governs all intelligent systems. We will trace how persistence against entropy drives the emergence of boundaries, sensing, learning, and world modelling across scales from the first living cells to human societies and future machine minds. The result is a single, scalable pattern: a framework for understanding how intelligence is structured, why it evolves, and where it is heading. We will conclude by looking ahead, asking what this architecture means for collective intelligence, for the next generation of machines, and for humanity’s own purpose in a world where intelligent systems will likely surpass us.Presenter: Max KellyMax Kelly is a strategist, entrepreneur and board director who has spent his career launching companies and shaping global strategy for organisations ranging from Virgin, Techstars and The Nature Conservancy to the University of Oxford’s Chemistry Department, while pursuing a deep fascination with how intelligence works.TNB Chair: Robert Chis-Ciure
Finding a Mathematical Language for Collective Intelligence - Jonas Hallgren Theoretical Neurobiology Group 2025-09-15 | As AI systems integrate into human organizations, we're seeing hybrid collective intelligence systems that combine market mechanisms, network effects, and hierarchical coordination simultaneously on the same social substrate. This talk addresses the mathematical challenge of extending active inference to multi-agent systems where different organizational forms operate at multiple scales while shaping each other's evolution over time. I propose using dynamic graph substrates combined with multi-scale coarse-graining to formalize how markets, networks, and hierarchies can emerge as different structural priors over the same Bayesian message-passing system, each optimizing collective free energy under distinct constraints. The framework treats different collective intelligence mechanisms as composable priors over evolving graph topologies where agents minimize prediction error at both individual and collective scales. This approach aims to simulate and predict hybrid human-AI collectives - from AI-augmented democratic processes to algorithmic markets embedded in social networks - extending the Free Energy Principle to collective systems.8th September 2025
High-dimensional Planning and Intrinsic Motivation - Tom Ringstrom Theoretical Neurobiology Group 2025-08-03 | Link for preprint: arxiv.org/pdf/2506.09499We introduce reward-free Option Kernel Bellman Equations (OKBEs). Rather than value functions, OKBEs directly construct and optimize a predictive map called a state-time option kernel (STOK) to maximize the probability of completing a goal while avoiding constraint violations. STOKs are compositional, modular, and interpretable initiation-to-termination transition kernels for policies in the Options Framework of Reinforcement Learning. This means: 1) STOKs can be sequentially composed using convolutional Chapman-Kolmogorov equations to make spatiotemporal predictions for multiple policies over long time-horizons, 2) high-dimensional STOKs can be represented and computed efficiently in a factorized and reconfigurable form, and 3) STOKs record the probabilities of semantically interpretable goal-success and constraint-violation events, needed for formal verification. Given a high-dimensional state-transition model for an intractable planning problem we can decompose it to optimize local STOKs and goal-conditioned policies that are aggregated into a factorized goal kernel, making it possible to forward-plan at the level of goals in a high-dimensional space to solve the original problem. These properties lead to highly flexible agents that can rapidly synthesize meta-policies, reuse planning representations for many tasks, and justify goals using empowerment, an intrinsic motivation function. We argue that reward-maximization is in conflict with the properties of compositionality, modularity, and interpretability. Alternatively, OKBEs facilitate these properties to support verifiable long-horizon planning and intrinsic motivation that can scale to dynamic high-dimensional world-models.Tuesday, February 4, 2025
Active Control of Internal Representations in SNNs and Robots - Alejandro Rodriguez Theoretical Neurobiology Group 2025-07-30 | Full Title: Movements on the Inside: Active Control of Internal Representations in Spiking Neural Networks and Robots Abstract: In this talk, I will discuss our recent work on active inference in two of the areas of expertise in our group. In the first one, we show a simple neural circuit capable of self-evidencing using continuous representations and the consequences of different architectural motifs and particular partitions. In the second one, we present a different kind of internal model used for human-robot or robot-robot interaction. We introduce the concept of informational coupling as means of reproducing dyadic dynamics observed in humans and animals. We conclude discussing our results in the light of their relations with some areas of physics and ecological psychology, presenting also some ongoing ideas and future work.Speaker Bio: Alejandro Jimenez Rodriguez is a senior lecturer in Artificial Intelligence and Robotics at Sheffield Hallam University. His work is at the interface of Computational Neuroscience and Robotics. He is particularly interested in autonomy, dynamical systems and self-organization.
How many flavours of surprise are there? by Héctor Manrique Theoretical Neurobiology Group 2025-07-23 | Abstract: Surprise lies at the heart of both Predictive Coding and Active Inference frameworks. Despite its conceptual centrality, it is sometimes conflated with novelty, and its multifaceted nature remains underexplored. I believe there may be multiple distinct "flavours" or subtypes of surprise yet to be discovered and/or described. The aim of this seminar is to investigate whether surprise can be systematically decomposed into qualitatively different types, and to examine the extent to which specific brain responses can be reliably associated with each of these types.Speaker Bio: I am a Professor of Developmental Psychology at the University of Zaragoza, Spain, and I approach the study of human cognition from an evolutionary-developmental (EvoDevo) perspective.
