Home
Artists
Search Recent Random
Posts
Search DMs Popular Hash Lookup Tags Random
Importer
Import FAQ
Register
Partychan Matrix ThePornDude
Home Artists Posts Import Register

BTS/WIP: Starting to explain Supervision Partially Observable Markov Decision Processes (Patreon)

Published:
2018-06-18 20:30:55
Imported:
2023-12
Tags:

Content

A preview of what I've been working on lately:
Turns out if you want to really explain this stuff fully, you have to start waaay back. Here's a draft lecture video we've been working on as part of the online course we're putting together. This one starts explaining stochastic processes and markov chains, building up to Supervision POMDPs. 

It's a little different from the main channel videos of course. A bit more formal. Anyway, any feedback is welcome!

Also, there's a big ol' reflection on the glass, which is annoying. Anyone know how to get rid of that? It's beyond my video effects ability right now.

https://www.youtube.com/watch?v=NgUG2O8dDOY

Files

POMDP Test

Comments

loopuleasa

Some philosophy here about reality. By the definitions provided, since the question popped in my head: Is reality Markovian? In the sense the future outcomes only depends on the present state. I guess so, given our understanding of physics. I may be missing something. I know brains are using past state information to generalize a predictive model that can look multiple states into the future, with some probabilities, but still the solid state of the brain and its memory and functions is a property of the local time moment of the environment, of the reality around the brain. Fascinating. Unsure of the time invariance property though, that might be trickier to explore as an idea, but from quantum mechanics, it seems time doesn't exist, so that might be respected too, based on that. These should be public ASAP, many people trying to study them would find them a godsend. I didn't understand the terms mentioned in the video before you explained them, although I am familiar with the underlying principles from probability, game theory, algorithmics and physics. I've increased my pledge. Quite impressed how easy you explained those concepts, and you somehow made the audience want to see the maths.

Jason Hise

'Present state of reality' is a confused concept once you are talking about either special or general relativity. Relative to a given point in space and time, 4D spacetime events can either exist in your past light cone, future light cone, or outside of both light cones. This tells you whether the event can have causally influenced you or whether you can causally influence it. But most of the points in spacetime are 'unordered' relative to each other because neither can influence the other. But if you are just talking about your past light cone that's probably fair; even if parts of the state in your past light cone are in quantum superposition that just makes parts of your current state probabilistic, which doesn't introduce any problems for computing probabilities of future states.