Shepherding Random Numbers
inconvergent.net
Shepherding Random Numbers
1–10 of 16 posts
Re: Shepherding Random Numbers
#2very nice animations
Re: Shepherding Random Numbers
#3Recommend following the link at the bottom: http://inconvergent.net/generative/
Very nice work. My favorite is this: http://inconvergent.net/generative/trees/
Re: Shepherding Random Numbers
#4Given how close to random walks (e.g., with absorption/reflection) some of these concoctions are, the probabilist in me wants a higher-level explanation, in a few cases, to be part of the post. Either way, it's fun to look at.
Re: Shepherding Random Numbers
#5As a mathematician but not a statistician or probabilist (so very much not an expert here), this:
> When we look at this it can sometimes seem as if there is a pattern emerging from the behaviour. However, if you feel like you can see such a pattern it is entirely coincidental, since the nodes are completely unaware of each other.
bothered me. It seems to me that the fact that patterns can emerge from non-coördinated behaviour is one of the interesting facts about mathematics. See, for a dodgy example—because there is some coördination—firefly flash-locking; and, for a genuine example that I literally just happened to be reading about this morning, the fascinating Nobel-Prize work discussed at https://johncarlosbaez.wordpress.com/2016/10/07/kosterlitz-t... . (I encourage you to go read the latter if you haven't already; as you will expect if you have read Baez's work, it's exposition at a level just about anyone, including non-mathematicians, can probably understand of scientific Nobel-Prize work, and those don't come along very often.) I guess, though, that to make any rigorous statement that this can occur, and isn't just an illusion, one has to make a rigorous definition of 'pattern', and the implicit definition of something like "externally imposed structure" has just as much claim to being 'correct' as anything I could cook up.
Re: Shepherding Random Numbers
#6That's amazing!
He also has a very interesting Twitter account where he posts his experiments: https://twitter.com/inconvergent
Re: Shepherding Random Numbers
#7I love his stuff. I've ported some of his code on GitHub to emit gcode and run it on my various drawing machines. Fun stuff.
Re: Shepherding Random Numbers
#8Yay stochastic differential equations..
Re: Shepherding Random Numbers
#9As a mathematician but not a statistician or probabilist (so very much not an expert here), this: > When we look at this it can sometimes seem as if there is a pattern emerging from the behaviour. However, if you feel like you can see such a pattern it is entirely coincidental, since the nodes are completely unaware of each other. bothered me. It seems to me that the fact that patterns can emerge from non-coördinated…
I'm pretty sure each point was rendered independently - i.e. the nodes have no interactions. So unlike flocking, they are are entirely coincidental.
Re: Shepherding Random Numbers
#10It's interesting how random numbers are essential in AI. Dropout and adding random Gaussian noise are standard practices. Random numbers are also essential in exploration strategies in Reinforcement Learning (epsilon greedy).
In the human brain, signals are stochastic, so they embed a large portion of randomness with each bit of information. Amazingly, adding noise to signals helps learning. Intelligence is at the edge of order and chaos - the lesson being that chaos is an essential part of it.