This is an excellent video on Active Inference. Easy to follow and quick intuitive. We're excited that more and more people are becoming aware of it! I'm a co-founder at Bioform Labs where we're building Active Inference tools for everyone to use. To date, we've developed a platform that uses Active Inference to build causal generative models of systems with measures of the model uncertainly (VFE) compared to real world observations (modalities). We can also decompose nodes in the models to reveal nested, cyclical causal structures and are developing EFE capabilities as I write this.
It seems to me that Dr. Chandaria is as excited about Karl Friston's FEP and Active Inference as I was when I learned about it by accident two years ago. It basically knocked my socks off at the time.
🎯 Key Takeaways for quick navigation: 00:01 🧠 Introduction and Importance of Predictive Processing - Introduction to the topics of the Bayesian brain, predictive processing, active inference, and the free energy principle. - Reference to philosophical aspects, particularly Kant's ideas on perception. - Introducing the concept of sensory data and how we form representations of the external world. 07:19 🤖 Machine Learning and Neural Networks as Inspiration - Discusses the role of artificial neural networks in training on sensory data. - Explains supervised learning in machine learning and its challenges. - Introduces the concept of autoencoders and variational autoencoders and their role in training on data. 15:32 🔑 Variational Inference and the Free Energy Principle - Describes how the variational algorithm works to solve approximate Bayesian inference. - Discusses the concept of KL Divergence and how it measures the informational distance between probability distributions. - Highlights the role of the free energy principle in training recognition and generative models, combining prediction error and prior expectations. 24:56 💡 Simplifying the Mathematics and Unraveling the Free Energy Principle - Encourages understanding the core ideas of the free energy principle beneath the technical mathematics. - Provides an example of calculating KL Divergence to show its simplicity. - Emphasizes the importance of grasping the fundamental concepts within the complexity of the theory. 26:58 🔍 Free Energy Principle and Variational Algorithms - Explanation of variational algorithms approximating posterior distributions. - Minimization of KL Divergence between approximated distribution and true posterior. - Incorporation of base theorem and separation of terms. 30:33 🧠 Unraveling the Free Energy Principle - The free energy concept and its connection to Helmoltz free energy. - The terms involved in free energy: surprise, KL Divergence, and more. - Exploring the concept of surprise in sensory data. 33:10 🌌 Phenomenal Self Model - Discussion of how sensory data and generative models construct conscious experience. - Introduction to the concept of a phenomenal self model. - Phenomenal transparency and the organization of conscious contents. 45:03 ⚙️ Active Inference and the Free Energy Principle - Introduction to active inference and its unification of action and perception. - Explanation of how action minimizes prediction error. - Integration of preferred outcomes and policies in active inference. 52:37 💭 Predictive Processing and Computational Neural Phenomenology - The ongoing debate about the relationship between predictive processing and consciousness. - The use of active inference frameworks to model phenomenal experience. - The potential role of posterior and precision weightings in modeling consciousness. 55:03 🤖 Framework or Theory of Predictive Processing - Discusses whether Predictive Processing is a framework or a theory. - Raises the need for Predictive Processing to explain phenomena like backward masking and binocular rivalry. - Mentions ongoing debates and the evolution of theories of consciousness. 59:09 🌍 Predictive Processing Beyond the Brain - Explores the idea that the Free Energy Principle and Predictive Processing go beyond the brain. - Introduces the concept of Markov blankets as informational boundaries. - Suggests that Markov blankets can be found not only in the brain but also in various systems, from cells to ecosystems. 01:01:57 🧩 Detecting Markov Blankets - Discusses the detection of Markov blankets in systems. - Highlights that most Markov blankets are approximate. - Explains that systems with Markov blankets are self-contained, persist through time, and have a model of their external environment. Made with HARPA AI
So, I take it you guys are doing some epistemologically subtle role out? Because we’re obviously living in something that is a kind of course grained VR representation of far more complex quantum phenomena. When this hit me, a few weeks ago, it was a “spiritual awakening” aka a disruptive paradigm shift from the external, objective, mind independent reality I believed I was in. It made me have a bit of a bad trip for 5-10 mins before I managed to construct a new framework along these lines. Anil seems to be gently seeding the zeitgeist? And Friston’s framework is so dense that it obfuscates the inevitable logical epistemological conclusions Fascinating stuff, but it has staggering implications for every aspect of science. What does it mean for cosmology and astrophysics if we’re studying a kind of low resolution interface that is obsessed with symmetry and best hypotheses? We’re not studying an external 3D structure
It’s amazing that the rabbit whole that lead to coming across these ideas starts at the last 2 chapters of Freud’s interpretation of dreams. After realizing how ahead of his time he was and how in the background to many psychologists dismay we really are completing his project
This is an excellent video on Active Inference. Easy to follow and quick intuitive. We're excited that more and more people are becoming aware of it!
I'm a co-founder at Bioform Labs where we're building Active Inference tools for everyone to use. To date, we've developed a platform that uses Active Inference to build causal generative models of systems with measures of the model uncertainly (VFE) compared to real world observations (modalities). We can also decompose nodes in the models to reveal nested, cyclical causal structures and are developing EFE capabilities as I write this.
This is very interesting. I am working on similar projects.
@@waylonbarrett3456 Let's connect!
It seems to me that Dr. Chandaria is as excited about Karl Friston's FEP and Active Inference as I was when I learned about it by accident two years ago. It basically knocked my socks off at the time.
Thanks for the video this was one of the best guides on free energy I’ve seen.
Bravo!
the rabbit hole of HHMM model leads me to markov blanket - > active inference - > free energy and now to this amazing tutorial
🎯 Key Takeaways for quick navigation:
00:01 🧠 Introduction and Importance of Predictive Processing
- Introduction to the topics of the Bayesian brain, predictive processing, active inference, and the free energy principle.
- Reference to philosophical aspects, particularly Kant's ideas on perception.
- Introducing the concept of sensory data and how we form representations of the external world.
07:19 🤖 Machine Learning and Neural Networks as Inspiration
- Discusses the role of artificial neural networks in training on sensory data.
- Explains supervised learning in machine learning and its challenges.
- Introduces the concept of autoencoders and variational autoencoders and their role in training on data.
15:32 🔑 Variational Inference and the Free Energy Principle
- Describes how the variational algorithm works to solve approximate Bayesian inference.
- Discusses the concept of KL Divergence and how it measures the informational distance between probability distributions.
- Highlights the role of the free energy principle in training recognition and generative models, combining prediction error and prior expectations.
24:56 💡 Simplifying the Mathematics and Unraveling the Free Energy Principle
- Encourages understanding the core ideas of the free energy principle beneath the technical mathematics.
- Provides an example of calculating KL Divergence to show its simplicity.
- Emphasizes the importance of grasping the fundamental concepts within the complexity of the theory.
26:58 🔍 Free Energy Principle and Variational Algorithms
- Explanation of variational algorithms approximating posterior distributions.
- Minimization of KL Divergence between approximated distribution and true posterior.
- Incorporation of base theorem and separation of terms.
30:33 🧠 Unraveling the Free Energy Principle
- The free energy concept and its connection to Helmoltz free energy.
- The terms involved in free energy: surprise, KL Divergence, and more.
- Exploring the concept of surprise in sensory data.
33:10 🌌 Phenomenal Self Model
- Discussion of how sensory data and generative models construct conscious experience.
- Introduction to the concept of a phenomenal self model.
- Phenomenal transparency and the organization of conscious contents.
45:03 ⚙️ Active Inference and the Free Energy Principle
- Introduction to active inference and its unification of action and perception.
- Explanation of how action minimizes prediction error.
- Integration of preferred outcomes and policies in active inference.
52:37 💭 Predictive Processing and Computational Neural Phenomenology
- The ongoing debate about the relationship between predictive processing and consciousness.
- The use of active inference frameworks to model phenomenal experience.
- The potential role of posterior and precision weightings in modeling consciousness.
55:03 🤖 Framework or Theory of Predictive Processing
- Discusses whether Predictive Processing is a framework or a theory.
- Raises the need for Predictive Processing to explain phenomena like backward masking and binocular rivalry.
- Mentions ongoing debates and the evolution of theories of consciousness.
59:09 🌍 Predictive Processing Beyond the Brain
- Explores the idea that the Free Energy Principle and Predictive Processing go beyond the brain.
- Introduces the concept of Markov blankets as informational boundaries.
- Suggests that Markov blankets can be found not only in the brain but also in various systems, from cells to ecosystems.
01:01:57 🧩 Detecting Markov Blankets
- Discusses the detection of Markov blankets in systems.
- Highlights that most Markov blankets are approximate.
- Explains that systems with Markov blankets are self-contained, persist through time, and have a model of their external environment.
Made with HARPA AI
This is a crazy good video …. Thnx a million.
Brilliant
Thank you for uploading, worth watching for anyone who is into free energy principle and AI
So, I take it you guys are doing some epistemologically subtle role out? Because we’re obviously living in something that is a kind of course grained VR representation of far more complex quantum phenomena. When this hit me, a few weeks ago, it was a “spiritual awakening” aka a disruptive paradigm shift from the external, objective, mind independent reality I believed I was in. It made me have a bit of a bad trip for 5-10 mins before I managed to construct a new framework along these lines.
Anil seems to be gently seeding the zeitgeist? And Friston’s framework is so dense that it obfuscates the inevitable logical epistemological conclusions
Fascinating stuff, but it has staggering implications for every aspect of science. What does it mean for cosmology and astrophysics if we’re studying a kind of low resolution interface that is obsessed with symmetry and best hypotheses? We’re not studying an external 3D structure
This is so complex. Why I like this? What will future humans do without mysteries? How can I compete with those asians?
The arguments of D_KL should be exchanged to Q || P at 28:00.
yes good spot, it should be D_KL(Q || P)! The RHS is correct , though, so it doesn't impact the derivation
@@ShamilChandaria Sure, just to avoid confusion (as you mentioned, D_KL is not a metric as it is not symmetric in its arguments).
Is the Action in your description of motion analogous to the Action in classical mechanics?
No, it‘s action in the usual sense of purposefully performing some change in the environment (in order to reduce surprise).
It’s amazing that the rabbit whole that lead to coming across these ideas starts at the last 2 chapters of Freud’s interpretation of dreams.
After realizing how ahead of his time he was and how in the background to many psychologists dismay we really are completing his project
What's the summary without using math?
brain makes up reality, but it's really good in doing so
Wut@@louis71
You're not making any sense, watch the video@@louis71
You are in a constant state of landing a plane.
I'd love to see this material divorced from the Aristotelean, Kantian and Cartesian baggage. Never going to happen, but I'd love to see it.
Aristotelean and Cartesian, certainly, but Kant is a prerequisite. Nay, a prior!