Uploaded August 2026 | Updated November 2026, 3 hours 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: the third of four sessions on Chapter 7
0:28 The 6-7-8 island: continuous time and what it buys you
1:01 Looking ahead to Chapters 9 and 10: discrete time and explicit counterfactuals
2:10 Q&A: reading the chapter dependency figure
3:52 Chapter 7 notes and the deferred topics
5:21 Why action is treated differently from hidden state inference
6:28 Page 161, Figure 7.2: the Bayesian network with v
7:31 Why there are no counterfactual actions in this formulation
8:07 The generative model and the generative process
8:36 7.1 Introducing v: from perceptual observer to active observer
9:36 Contrast with reinforcement learning policy rollouts
10:44 7.2 and Equation 7.5: action as a free energy gradient
11:38 Side note 9: the action learning rate as inertia
12:40 Euler stepping: action at the next time step
13:46 Action as something continuously unfolding, not an event
15:01 The problem: free energy is not a function of action
15:41 The workaround: actions have sensory consequences
17:08 Applying the chain rule (7.5 to 7.7)
18:38 Splitting the fraction into two interpretable terms
19:39 Domain-specific sensory change vs the free energy objective
20:35 Equation 7.8: the derivative of free energy w.r.t. sensory data
22:28 Equation 7.9 and its parallel to 7.5
23:29 Equation 7.12: forward model, sensory precision, prediction error
24:41 Equations 7.10a and b: change your mind and change the world
25:26 Perception and action under a single objective
26:07 Why the indirect route through observations is the right one
28:32 Page 163: the procedural loop
29:49 Q&A: active inference for cognitive modelling and interoception
31:50 Computational psychiatry, model fitting, and simulating people
34:54 Unified models, hormones, and quantifying patchy data
36:41 The rub: how does the forward model get learned?
37:18 The three light bulbs riddle
38:47 Structure learning and popping up a level
40:12 Figure 7.3: internal and active states minimising the same free energy
41:13 The true external states and their noise terms
42:26 The Markov blanket property, precisely stated
43:05 The Markov blanket hype cycle
43:55 7.4 The three examples
45:14 7.5 Multivariate active inference in generalized coordinates
46:07 7.6 Conclusions: eye movement, motor control, reflex arcs
46:39 Why continuous time suits embodied modelling
47:21 Active inference does not contradict folk psychology
49:06 Chapter 7 in perspective: it builds on everything before it
49:47 Q&A: how to handle null and missing values
52:22 Parsing events, and why context is the dynamic part
54:37 Missing data in practice: imputation and domain judgement
56:29 Closing: affect labelling and communicative syntax/_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: the third of four sessions on Chapter 7
0:28 The 6-7-8 island: continuous time and what it buys you
1:01 Looking ahead to Chapters 9 and 10: discrete time and explicit counterfactuals
2:10 Q&A: reading the chapter dependency figure
3:52 Chapter 7 notes and the deferred topics
5:21 Why action is treated differently from hidden state inference
6:28 Page 161, Figure 7.2: the Bayesian network with v
7:31 Why there are no counterfactual actions in this formulation
8:07 The generative model and the generative process
8:36 7.1 Introducing v: from perceptual observer to active observer
9:36 Contrast with reinforcement learning policy rollouts
10:44 7.2 and Equation 7.5: action as a free energy gradient
11:38 Side note 9: the action learning rate as inertia
12:40 Euler stepping: action at the next time step
13:46 Action as something continuously unfolding, not an event
15:01 The problem: free energy is not a function of action
15:41 The workaround: actions have sensory consequences
17:08 Applying the chain rule (7.5 to 7.7)
18:38 Splitting the fraction into two interpretable terms
19:39 Domain-specific sensory change vs the free energy objective
20:35 Equation 7.8: the derivative of free energy w.r.t. sensory data
22:28 Equation 7.9 and its parallel to 7.5
23:29 Equation 7.12: forward model, sensory precision, prediction error
24:41 Equations 7.10a and b: change your mind and change the world
25:26 Perception and action under a single objective
26:07 Why the indirect route through observations is the right one
28:32 Page 163: the procedural loop
29:49 Q&A: active inference for cognitive modelling and interoception
31:50 Computational psychiatry, model fitting, and simulating people
34:54 Unified models, hormones, and quantifying patchy data
36:41 The rub: how does the forward model get learned?
37:18 The three light bulbs riddle
38:47 Structure learning and popping up a level
40:12 Figure 7.3: internal and active states minimising the same free energy
41:13 The true external states and their noise terms
42:26 The Markov blanket property, precisely stated
43:05 The Markov blanket hype cycle
43:55 7.4 The three examples
45:14 7.5 Multivariate active inference in generalized coordinates
46:07 7.6 Conclusions: eye movement, motor control, reflex arcs
46:39 Why continuous time suits embodied modelling
47:21 Active inference does not contradict folk psychology
49:06 Chapter 7 in perspective: it builds on everything before it
49:47 Q&A: how to handle null and missing values
52:22 Parsing events, and why context is the dynamic part
54:37 Missing data in practice: imputation and domain judgement
56:29 Closing: affect labelling and communicative syntax/_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/








![ActInf GuestStream 115.1 ~ Energy-Based Transformers and the Future of Scaling
Energy-Based Transformers are Scalable Learners and Thinkers
Alexi Gladstone, Ganesh Nanduru, Md Mofijul Islam, Peixuan Han, Hyeonjeong Ha, Aman Chadha, Yilun Du, Heng Ji, Jundong Li, Tariq Iqbal
https://arxiv.org/abs/2507.02092
Inference-time computation techniques, analogous to human System 2 Thinking, have recently become popular for improving model performances. However, most existing approaches suffer from several limitations: they are modality-specific (e.g., working only in text), problem-specific (e.g., verifiable domains like math and coding), or require additional supervision/training on top of unsupervised pretraining (e.g., verifiers or verifiable rewards). In this paper, we ask the question Is it possible to generalize these System 2 Thinking approaches, and develop models that learn to think solely from unsupervised learning? Interestingly, we find the answer is yes, by learning to explicitly verify the compatibility between inputs and candidate-predictions, and then re-framing prediction problems as optimization with respect to this verifier. Specifically, we train Energy-Based Transformers (EBTs) a new class of Energy-Based Models (EBMs) to assign an energy value to every input and candidate-prediction pair, enabling predictions through gradient descent-based energy minimization until convergence. Across both discrete (text) and continuous (visual) modalities, we find EBTs scale faster than the dominant Transformer++ approach during training, achieving an up to 35% higher scaling rate with respect to data, batch size, parameters, FLOPs, and depth. During inference, EBTs improve performance with System 2 Thinking by 29% more than the Transformer++ on language tasks, and EBTs outperform Diffusion Transformers on image denoising while using fewer forward passes. Further, we find that EBTs achieve better results than existing models on most downstream tasks given the same or worse pretraining performance, suggesting that EBTs generalize better than existing approaches. Consequently, EBTs are a promising new paradigm for scaling both the learning and thinking capabilities of models.
Subjects: Machine Learning (cs.LG); Artificial Intelligence (cs.AI); Computation and Language (cs.CL); Computer Vision and Pattern Recognition (cs.CV)
Cite as: arXiv:2507.02092 [cs.LG]
(or arXiv:2507.02092v1 [cs.LG] for this version)
https://doi.org/10.48550/arXiv.2507.02092
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 115.1 ~ Energy-Based Transformers and the Future of Scaling](https://i.ytimg.com/vi/85ob-jrG7xQ/mqdefault.jpg)

