Active inference explained with Prof. Karl Friston
ฝัง
- เผยแพร่เมื่อ 30 ก.ค. 2024
- 📢📢📢 We focus on active inference explained by Prof. Karl Friston the Godfather of Neuroscience. We will also cover: Perceptual Inference, Over-parametrized models, active inference vs. reinforcement learning, etc. Many thanks to Prof. Jamshid Ghasimi ( / jamshid-ghasimi-992a30145 ) for helping me design high-quality questions.
----------------------------------------------------------------------------------------------------------------------------------
00:00 A brief introduction of Prof. Friston
00:51 What is the Free-Energy Principle?
09:13 Model evidence optimization (can brain OVERFIT?!) and how BRAIN selects the actions with minimum expected surprise!
28:23 The double-descent phenomenon in over-parametrized models (Including our brain) and WHY these models DO NOT overfit?
40:37 Active Inferencs vs. Perceptual Inference and how to they work together for US to LEARN!
52:10 The difference between Active Inference and Reinforcement Learning
01:08:32 How 'Abstract Knowledge' generation (Creativity) in brain can be explained?
01:19:45 The ARCHITECTURE of your Nervous System describes the ENVIRONMENT you are living in!
01:31:48 Will we ever FULLY understand how the brain works and how far have we come?
01:35:00 The Goodbye
----------------------------------------------------------------------------------------------------------------------------------
🔴 The GitHub loink to the project: github.com/MLDawn/MLDawn-Proj...
🌎 You can visit our Website: www.mldawn.com/
🕊 You can follow us on Twitter: / mldawn2018
🔗 You can join us on Linked In: / mehran-ba. .
🔗 You can join us on Facebook: / ml.dawn.3
🔗 You can join us on Instagram: mldawn2018
Keep up the good work and good luck! 🤞🍀
This is one of the most profound and informative dialogue with Karl Friston..deeply grateful for your stimulating questions
Thanks a lot. Many thanks goes to Prof. Jamshid Ghasimi (www.linkedin.com/in/jamshid-ghasimi-992a30145/) who helped me design high-quality questions that were appropriate to a guest of Prof. Friston's stature. I am indeed grateful to Prof. Friston for the opportunity.
My dear son
Congradulation, great job
Narges
Thanks a lot for all your support
Ive been watching every interview with friston and this is one of the if not the best one.
Thank you very much for this amazing interview about the most importand subject
You are most welcome. It was such an honor to have Prof. Friston as my guest
35:38 beautiful metaphor of the "mountain range" (genesis of stream becoming a river as it goes downhill). Another lay-person here, loving every time a real life example is made. I second @WP Thank you for posting this!!!
I know, right? Prof. Friston describes all of these technical concepts using the most simplest analogies. I am happy you are enjoying the video.
Great video, as always, keep it up MLDawn
Appreciate it
Thank you for a genuinely delightful discussion. I am a lay person in all the subject matter that relates to the Free Energy Principle (FEP), whether it is physics, machine learning, robotics, thermodynamics, information science, or what have you. I am exploring the topic in an attempt to grasp where science has brought us in answering those age old questions about being, self, free will, etc., that philosophers and theologians have pondered for millennia. Each time I see one of these interdisciplinary discussions with Professor Friston I get new perspectives and improved understanding of the science and the philosophy. In this case, the discussion about over parameterizations finely cleared up in my mind Friston's reference to "sharp minima," which I had heard him mention before. The part discussing the difference between structured learning and FEP was similarly informative. Very well done on those topics and throughout.
I truly appreciate you taking the time to write down a feedback. Honestly, it was one of the unforgettable experiences of my life.
Excellent, brother, excellent! Both the content and the idea of such interviews :)
I really appreciate it
It would be easier to look at the equations after watching the human talk. My beliefs about active inference are starting to form.
I'm developing an artificial agent based on active inference and epistemic foraging. If anyone would like to help or offer suggestions, I'd like to hear from you.
Great stuff! I would suggest the simple T-maze scenario for your initial development with discrete states and actions. Then you can make it more sophisticated.
@@MLDawn That's a great idea. Thanks. Would you be interested in offering a critique of my model's code in its current state? Or, do you know of anyone looking to collaborate on such a project?