Uploaded July 2026 | Updated November 2026, 1 day ago
Book: Fundamentals of Active Inference
Principles, Algorithms, and Applications of the Free Energy Principle for Engineers, by Sanjeev V. Namjoshi https://mitpress.mit.edu/9780262050951/fundamentals-of-active-inference/
Register: http://textbook-group.activeinference.institute/
Code: github.com/ActiveInferenceInstitute/fundamentals
Playlist of Textbook Group: youtube.com/playlist?list=PLNm0u2n1IwdrpoTg-9l37YIBiDrrOzTw-
Direct link to group document: coda.io/d/_dZHp_UGhVZE/Meetings_su3ToqMp
------
CHAPTERS
0:00 Intro: Chapter 7, the watershed chapter where action enters
1:03 Q&A: MATLAB or Python for implementing hierarchical active inference?
4:15 SPM, DEM and the legacy MATLAB codebase
7:28 Translating SPM to Python: the pitfalls of LLM-assisted porting
8:33 The RGM paper and the translated subset
13:59 Multivariate active generalized filtering
15:03 7.1 Starting point: what is new in Chapter 7
17:10 Exogenous force (v*) and action (a)
18:11 Q&A: what does exogenous force mean, with an example
19:14 The Markov blanket as the agent-environment interface
24:26 Action vs exogenous force: two inputs to the generative process
27:36 Why there is no direct line between agent and environment
28:41 An observation is the environment's action
29:42 Why we build artificial environments for simulation
31:46 7.5 Putting action into the generative model
32:52 Observations as functions of action
33:56 The free energy gradient with respect to observations
35:05 The duality of action and perception
36:06 Equations 7.11 a-d
37:07 The motion of action: learning rate, forward model, precision, prediction error
38:12 Example 7.1: the hydra in the water column
39:16 Specifying the environment and the observation function
40:20 The autonomous state as a preference
41:27 Figure 7.5: what happens with perception only
43:33 Observations, beliefs and actions on one plot
44:35 Free energy before and after action is enabled
45:16 Example 7.2: adding an exogenous force
46:12 Figure 7.8: results with an exogenous input
47:36 Two ways to resolve prediction error
48:00 Sensory surprise over time: perceive only vs perceive and act
49:11 Figure 7.11: why perception alone cannot reduce surprisal
50:33 Example 7.3: countering a dynamic exogenous force
51:55 The bird on the branch
52:31 7.5 Multivariate active inference and Algorithm 11
53:05 Plan for next session: the Bayesian thermostat in code
53:55 Andrew on Example 7.3 and the supplementary slides
56:07 Q&A: v* vs v, homeostatic set points and preference priors
58:44 Figure 7.3 again: the agent-environment interface and its duality
59:47 Q&A: dynamic environments and self-driving cars
1:03:36 Active inference vs standard machine learning losses
1:07:05 Renormalising generative models and object identification
1:10:03 When the environment is made of other agents: theory of mind
1:11:12 Q&A: is the exogenous force internal or external?
1:12:33 Perspectivism: everything outside agent one is environment
1:13:56 Markov blankets: delimiting the agent from the environment
1:17:04 Q&A: unmapped situations and missing context
1:19:33 Univariate vs multivariate: state space size and compute limits
1:20:53 Asynchronous fallbacks for autonomous systems
1:22:00 Wrap-up and next session/_dZHp_UGhVZE/Meetings_su3ToqMp
------
Active Inference Institute information:
Website: https://activeinference.institute/
X: https://x.com/InferenceActive
Discord: https://discord.activeinference.institute/
Donate: https://donate.activeinference.institute/
YouTube: youtube.com/c/ActiveInference
Active Inference Livestreams: https://video.activeinference.institute/
Book: Fundamentals of Active Inference
Principles, Algorithms, and Applications of the Free Energy Principle for Engineers, by Sanjeev V. Namjoshi https://mitpress.mit.edu/9780262050951/fundamentals-of-active-inference/
Register: http://textbook-group.activeinference.institute/
Code: github.com/ActiveInferenceInstitute/fundamentals
Playlist of Textbook Group: youtube.com/playlist?list=PLNm0u2n1IwdrpoTg-9l37YIBiDrrOzTw-
Direct link to group document: coda.io/d/_dZHp_UGhVZE/Meetings_su3ToqMp
------
CHAPTERS
0:00 Intro: Chapter 7, the watershed chapter where action enters
1:03 Q&A: MATLAB or Python for implementing hierarchical active inference?
4:15 SPM, DEM and the legacy MATLAB codebase
7:28 Translating SPM to Python: the pitfalls of LLM-assisted porting
8:33 The RGM paper and the translated subset
13:59 Multivariate active generalized filtering
15:03 7.1 Starting point: what is new in Chapter 7
17:10 Exogenous force (v*) and action (a)
18:11 Q&A: what does exogenous force mean, with an example
19:14 The Markov blanket as the agent-environment interface
24:26 Action vs exogenous force: two inputs to the generative process
27:36 Why there is no direct line between agent and environment
28:41 An observation is the environment's action
29:42 Why we build artificial environments for simulation
31:46 7.5 Putting action into the generative model
32:52 Observations as functions of action
33:56 The free energy gradient with respect to observations
35:05 The duality of action and perception
36:06 Equations 7.11 a-d
37:07 The motion of action: learning rate, forward model, precision, prediction error
38:12 Example 7.1: the hydra in the water column
39:16 Specifying the environment and the observation function
40:20 The autonomous state as a preference
41:27 Figure 7.5: what happens with perception only
43:33 Observations, beliefs and actions on one plot
44:35 Free energy before and after action is enabled
45:16 Example 7.2: adding an exogenous force
46:12 Figure 7.8: results with an exogenous input
47:36 Two ways to resolve prediction error
48:00 Sensory surprise over time: perceive only vs perceive and act
49:11 Figure 7.11: why perception alone cannot reduce surprisal
50:33 Example 7.3: countering a dynamic exogenous force
51:55 The bird on the branch
52:31 7.5 Multivariate active inference and Algorithm 11
53:05 Plan for next session: the Bayesian thermostat in code
53:55 Andrew on Example 7.3 and the supplementary slides
56:07 Q&A: v* vs v, homeostatic set points and preference priors
58:44 Figure 7.3 again: the agent-environment interface and its duality
59:47 Q&A: dynamic environments and self-driving cars
1:03:36 Active inference vs standard machine learning losses
1:07:05 Renormalising generative models and object identification
1:10:03 When the environment is made of other agents: theory of mind
1:11:12 Q&A: is the exogenous force internal or external?
1:12:33 Perspectivism: everything outside agent one is environment
1:13:56 Markov blankets: delimiting the agent from the environment
1:17:04 Q&A: unmapped situations and missing context
1:19:33 Univariate vs multivariate: state space size and compute limits
1:20:53 Asynchronous fallbacks for autonomous systems
1:22:00 Wrap-up and next session/_dZHp_UGhVZE/Meetings_su3ToqMp
------
Active Inference Institute information:
Website: https://activeinference.institute/
X: https://x.com/InferenceActive
Discord: https://discord.activeinference.institute/
Donate: https://donate.activeinference.institute/
YouTube: youtube.com/c/ActiveInference
Active Inference Livestreams: https://video.activeinference.institute/










