Uploaded June 2026 | Updated November 2026, 3 days ago
Online Generalised Predictive Coding
Mehran H. Z. Bazargani, Szymon Urbas, Adeel Razi, Thomas Brendan Murphy, Karl Friston
arxiv.org/abs/2605.02675
This paper introduces an extension of generalised filtering for online applications. Generalised filtering refers to data assimilation schemes that jointly infer latent states, learn unknown model parameters, and estimate uncertainty in an integrated framework -- e.g., estimate state and observation noise -- at the same time (i.e., triple estimation). This framework appears across disciplines under different names, including variational Kalman-Bucy filtering in engineering, generalised predictive coding in neuroscience, and Dynamic Expectation Maximisation (DEM) in time-series analysis. Here, we specialise DEM for ``online'' data assimilation, through a separation of temporal scales. We describe the variational principles and procedures that allow one to assimilate data in a way that allows for a slow updating of parameters and precisions, which contextualise fast Bayesian belief updating about the dynamic hidden states. Using numerical studies, we demonstrate the validity of online DEM (ODEM) using a non-linear -- and potentially chaotic -- generative model, to show that the ODEM scheme can track the latent states of the generative process, even when its functional form differs fundamentally from the dynamics of the generative model. Framed from a neuro-mimetic predictive coding perspective, ODEM offers a biologically inspired solution to online inference, learning, and uncertainty estimation in dynamic environments.
doi.org/10.48550/arXiv.2605.02675
Code: github.com/mldawn/odem
-
--- RESOURCES & TRANSCRIPT ---
đ Full Transcript on GitHub:
github.com/ActiveInferenceInstitute/ActiveInferenceJournal/tree/main/data/video/activeinferenceinstitute/ModelStream/ModelStream_008
Active Inference Institute information:
Website: https://www.activeinference.institute/
Activities: https://activities.activeinference.institute/
Discord: https://discord.activeinference.institute/
Donate: http://donate.activeinference.institute/
YouTube: youtube.com/c/ActiveInference
X: https://x.com/InferenceActive
Active Inference Livestreams: https://video.activeinference.institute/
------
CHAPTERS
0:00 Welcome and introduction
1:30 The brain as an inference machine
3:00 Generative models and generative processes
4:30 Markov blankets
6:00 Temporal derivatives and dynamics
7:30 State and observation equations
9:00 Smooth Gaussian noise
10:30 Unknowns to infer
12:00 Generalized coordinates of motion
13:30 The derivative operator
16:30 Variational free energy
18:00 Variational inference
21:00 Gaussian assumptions
22:30 Approximating the posterior
25:30 Divergence from priors
27:00 Predictive coding vs. active inference
28:30 Hierarchical predictive coding
30:00 Local learning
31:30 Passing prediction errors
33:00 Prediction errors on dynamics
34:30 Gradient descent in generalized coordinates
36:00 Dynamics as momentum
37:30 Separation of time scales
40:30 The online algorithm
42:00 Accumulating gradients
43:30 Free action
45:00 Relation to dynamic causal modeling
46:30 E-step and M-step loops
48:00 Python implementation
49:30 Experimental setup
51:00 Generalized Lotka-Volterra system
52:30 Generating observations
55:30 Two scenarios: matched vs. mismatched dynamics
57:00 Observation mapping
58:30 Priors and the interaction matrix
1:00:00 Precision ratio as attention
1:01:30 Orders of motion
1:04:30 Results: the precision ratio
1:09:00 Trade-off: noisy estimates
1:13:30 Results: orders of motion
1:16:30 Parameter and noise estimates
1:22:30 No ground truth for parameters
1:24:00 Future work: hierarchical models
1:25:30 Dimensionality reduction
1:27:00 Acknowledgments
1:28:30 Code walkthrough: package architecture
1:30:00 Computing variational free energy
1:31:30 Running an experiment
1:34:30 Time-scale separation in code
1:36:00 Defining the generative model
1:37:30 Outputs
1:40:30 Plotting results
1:42:00 Q&A: comparing orders of motion
1:45:00 Q&A: no domain knowledge required
1:46:30 Closing thoughts
Online Generalised Predictive Coding
Mehran H. Z. Bazargani, Szymon Urbas, Adeel Razi, Thomas Brendan Murphy, Karl Friston
arxiv.org/abs/2605.02675
This paper introduces an extension of generalised filtering for online applications. Generalised filtering refers to data assimilation schemes that jointly infer latent states, learn unknown model parameters, and estimate uncertainty in an integrated framework -- e.g., estimate state and observation noise -- at the same time (i.e., triple estimation). This framework appears across disciplines under different names, including variational Kalman-Bucy filtering in engineering, generalised predictive coding in neuroscience, and Dynamic Expectation Maximisation (DEM) in time-series analysis. Here, we specialise DEM for ``online'' data assimilation, through a separation of temporal scales. We describe the variational principles and procedures that allow one to assimilate data in a way that allows for a slow updating of parameters and precisions, which contextualise fast Bayesian belief updating about the dynamic hidden states. Using numerical studies, we demonstrate the validity of online DEM (ODEM) using a non-linear -- and potentially chaotic -- generative model, to show that the ODEM scheme can track the latent states of the generative process, even when its functional form differs fundamentally from the dynamics of the generative model. Framed from a neuro-mimetic predictive coding perspective, ODEM offers a biologically inspired solution to online inference, learning, and uncertainty estimation in dynamic environments.
doi.org/10.48550/arXiv.2605.02675
Code: github.com/mldawn/odem
-
--- RESOURCES & TRANSCRIPT ---
đ Full Transcript on GitHub:
github.com/ActiveInferenceInstitute/ActiveInferenceJournal/tree/main/data/video/activeinferenceinstitute/ModelStream/ModelStream_008
Active Inference Institute information:
Website: https://www.activeinference.institute/
Activities: https://activities.activeinference.institute/
Discord: https://discord.activeinference.institute/
Donate: http://donate.activeinference.institute/
YouTube: youtube.com/c/ActiveInference
X: https://x.com/InferenceActive
Active Inference Livestreams: https://video.activeinference.institute/
------
CHAPTERS
0:00 Welcome and introduction
1:30 The brain as an inference machine
3:00 Generative models and generative processes
4:30 Markov blankets
6:00 Temporal derivatives and dynamics
7:30 State and observation equations
9:00 Smooth Gaussian noise
10:30 Unknowns to infer
12:00 Generalized coordinates of motion
13:30 The derivative operator
16:30 Variational free energy
18:00 Variational inference
21:00 Gaussian assumptions
22:30 Approximating the posterior
25:30 Divergence from priors
27:00 Predictive coding vs. active inference
28:30 Hierarchical predictive coding
30:00 Local learning
31:30 Passing prediction errors
33:00 Prediction errors on dynamics
34:30 Gradient descent in generalized coordinates
36:00 Dynamics as momentum
37:30 Separation of time scales
40:30 The online algorithm
42:00 Accumulating gradients
43:30 Free action
45:00 Relation to dynamic causal modeling
46:30 E-step and M-step loops
48:00 Python implementation
49:30 Experimental setup
51:00 Generalized Lotka-Volterra system
52:30 Generating observations
55:30 Two scenarios: matched vs. mismatched dynamics
57:00 Observation mapping
58:30 Priors and the interaction matrix
1:00:00 Precision ratio as attention
1:01:30 Orders of motion
1:04:30 Results: the precision ratio
1:09:00 Trade-off: noisy estimates
1:13:30 Results: orders of motion
1:16:30 Parameter and noise estimates
1:22:30 No ground truth for parameters
1:24:00 Future work: hierarchical models
1:25:30 Dimensionality reduction
1:27:00 Acknowledgments
1:28:30 Code walkthrough: package architecture
1:30:00 Computing variational free energy
1:31:30 Running an experiment
1:34:30 Time-scale separation in code
1:36:00 Defining the generative model
1:37:30 Outputs
1:40:30 Plotting results
1:42:00 Q&A: comparing orders of motion
1:45:00 Q&A: no domain knowledge required
1:46:30 Closing thoughts
![ActInf GuestStream 118.1 ~ Tadahiro Taniguchi Co-Creative Learning via Metropolis-Hastings...
Co-Creative Learning via Metropolis-Hastings Interaction between Humans and AI
Ryota Okumura, Tadahiro Taniguchi, Akira Taniguchi, Yoshinobu Hagiwara
https://arxiv.org/abs/2506.15468
[Submitted on 18 Jun 2025]
We propose co-creative learning as a novel paradigm where humans and AI, i.e., biological and artificial agents, mutually integrate their partial perceptual information and knowledge to construct shared external representations, a process we interpret as symbol emergence. Unlike traditional AI teaching based on unilateral knowledge transfer, this addresses the challenge of integrating information from inherently different modalities. We empirically test this framework using a human-AI interaction model based on the Metropolis-Hastings naming game (MHNG), a decentralized Bayesian inference mechanism. In an online experiment, 69 participants played a joint attention naming game (JA-NG) with one of three computer agent types (MH-based, always-accept, or always-reject) under partial observability. Results show that human-AI pairs with an MH-based agent significantly improved categorization accuracy through interaction and achieved stronger convergence toward a shared sign system. Furthermore, human acceptance behavior aligned closely with the MH-derived acceptance probability. These findings provide the first empirical evidence for co-creative learning emerging in human-AI dyads via MHNG-based interaction. This suggests a promising path toward symbiotic AI systems that learn with humans, rather than from them, by dynamically aligning perceptual experiences, opening a new venue for symbiotic AI alignment.
GuestStream #104.1 https://www.youtube.com/live/veGQUGE0hLI Tadahiro Taniguchi : Collective Predictive Coding and Active Inference: A New Perspective on Emergent Communication
Active Inference Institute information:
Website: https://www.activeinference.institute/
Activities: https://activities.activeinference.institute/
Discord: https://discord.activeinference.institute/
Donate: http://donate.activeinference.institute/
YouTube: https://www.youtube.com/c/ActiveInference/
X: https://x.com/InferenceActive
Active Inference Livestreams: https://video.activeinference.institute/ ActInf GuestStream 118.1 ~ Tadahiro Taniguchi Co-Creative Learning via Metropolis-Hastings...](https://i.ytimg.com/vi/0t14D4epcuw/mqdefault.jpg)




![Active Inference Working Groups: âCoordinating the Fieldâ
Prerecorded overview for the Active Inference Working Groups sessions at the 6th Applied Active Inference Symposium (November 12â13, 2026), hosted by the Active Inference Institute. Online and free.
How can a growing research field learn together? This video walks through the position paper âCoordinating the Field: The Case for Ecosystem-Level Technology Roadmapping and Community Working Groups in Active Inferenceâ (Ober, Maren, Friston, Hurtado, Pashea, Friedman, Sabine & Safron, 2026; AII Working Paper Series v1.0.0), together with the draft Working Group Charter and Operating Procedures.
The Active Inference Institute proposes open, chartered community working groups: one Field Working Group for field-level work (shared terminology, comparison across implementations, a shared assessment framework) and domain working groups that adapt and assess Active Inference in specific application areas. Membership is open to anyone with relevant expertise or substantive interest, including critics of Active Inference and domain practitioners who are not proponents of the framework. The draft documents carry no force until adopted.
Chapters:
0:00 Why coordinate?
2:30 What the evidence supports
5:00 The Field Working Group
7:25 Domain groups and literacy
9:48 A shared developmental framework
12:26 Governance that earns trust
15:04 A bounded pilot
17:43 From interest to contribution
Working Groups sessions at the Symposium (led by Ed Ober; open to all registered participants, with no obligation to join a group):
Session 1, Opening & Charter: Thursday 12 November, 16:00â17:00 UTC (Day 1 livestream)
Session 2, Outcomes & Next Steps: Friday 13 November, 20:30â21:30 UTC (Day 2 livestream)
Session details: https://docs.superhuman.com/d/6th-Applied-Active-Inference-Symposium-Public-Program_dfDPUsAuSgQ/Additional-Session-Information_suLKhYbG
Read and take part:
Position paper (Zenodo): https://doi.org/10.5281/zenodo.23066904
Draft Working Group Charter (v1.0.0): https://github.com/ActiveInferenceInstitute/Working_Groups/blob/main/docs/working_group_charter.md
Draft Operating Procedures (v1.0.0): https://github.com/ActiveInferenceInstitute/Working_Groups/blob/main/docs/operating_procedures.md
Comment on the drafts or express interest in a group (GitHub issues): https://github.com/ActiveInferenceInstitute/Working_Groups/issues
Working Groups overview: https://activeinference.institute/working-groups/
Register for the Symposium (free, online): https://registersymposium2026.activeinference.institute/
Email: blanket@activeinference.institute (subject line [WORKING GROUPS])
Symposium program: https://2026symposium.activeinference.institute/
Abstract book (Zenodo): https://zenodo.org/records/23986505
Livestreams: Day 1 https://www.youtube.com/watch?v v4IaFY9MZI · Day 2 https://www.youtube.com/watch?v=ccze9ay8D0o
All Symposium videos (playlist): https://www.youtube.com/playlist?list=PLO2noumu0LRk
Previous Symposia: https://symposium.activeinference.institute/
Active Inference Institute information:
Website: https://activeinference.institute/
X: https://x.com/InferenceActive
Discord: https://discord.activeinference.institute/
Donate: https://donate.activeinference.institute/
YouTube: https://www.youtube.com/c/ActiveInference/
Active Inference Livestreams: https://video.activeinference.institute/ Active Inference Working Groups: âCoordinating the Fieldâ](https://i.ytimg.com/vi/2JV08bgyJjg/mqdefault.jpg)
![ActInf ModelStream 021.1: Melih Kandemir: Distributional Active Inference
[Submitted on 28 Jan 2026]
Distributional Active Inference
Abdullah AkgĂŒl, Gulcin Baykal, Manuel HauĂmann, Mustafa Mert Ăelikok, Melih Kandemir
https://arxiv.org/abs/2601.20985
Code: https://github.com/adinlab/objectrl
Optimal control of complex environments with robotic systems faces two complementary and intertwined challenges: efficient organization of sensory state information and far-sighted action planning. Because the reinforcement learning framework addresses only the latter, it tends to deliver sample-inefficient solutions. Active inference is the state-of-the-art process theory that explains how biological brains handle this dual problem. However, its applications to artificial intelligence have thus far been limited to extensions of existing model-based approaches. We present a formal abstraction of reinforcement learning algorithms that spans model-based, distributional, and model-free approaches. This abstraction seamlessly integrates active inference into the distributional reinforcement learning framework, making its performance advantages accessible without transition dynamics modeling.
Active Inference Institute information:
Website: https://www.activeinference.institute/
Activities: https://activities.activeinference.institute/
Discord: https://discord.activeinference.institute/
Donate: http://donate.activeinference.institute/
YouTube: https://www.youtube.com/c/ActiveInference/
X: https://x.com/InferenceActive
Active Inference Livestreams: https://video.activeinference.institute/ ActInf ModelStream 021.1: Melih Kandemir: Distributional Active Inference](https://i.ytimg.com/vi/2NB7hTsp_F0/mqdefault.jpg)



