Radix Octrees

As I understand, they can be used to predict the next state of a system: think using them with Markov Chains.

I.e. we use markov chain on every pixel, then integrate the state to then next octree level, calculate again, and so on, till we reach tree root, having the folded prediction.

Say you write an AI for an RTS game. The current game map contains AI player base with enemy units near it. AI has memory from previous plays, which got him octree that enemy units near his base with have high probability following attack on it's base, which would trigger other octree, having memory that moving his AI's own units near base, would efficiently counter the attack. And so AI does that.

Basically you can use Radix Octrees with some "glue" code to completely encode AI behaviour.

Sounds like Mentifex's theory of mind, except that instead of five senses you're only using sight.

>>2

Nope. I'm using map state + optional fog of war.

>>3
Are you really thinking of using raw pixel planes as input, or was that just a (very bad) example?

Because that's retarded. That's like learning to read by studying the molecular makeup of ink.

>>4

Are you really thinking of using raw pixel planes as input, or was that just a (very bad) example?

AFAIK, animal brain does some filtering, like edge detection and feature extraction, before starting pattern matching. But in our case of a chess-like video game with rock-paper-scissors mechanics we can simplify processing by using pixels, say red pixel for enemy unit, green for friendly, and blue for base.

Because that's retarded. That's like learning to read by studying the molecular makeup of ink.

Octrees would aggregate your "ink" into letters and letters into words. So molecules of ink, like a red point (enemy unit) few pixels away from a base (blue point), would end up being letters (a state), which could be into a markov chain or a radix tree to get prediction/response.

>>5

could fed be into a markov chain or a radix tree to get prediction/response.

self fix

so given letters:
e - enemy unit
b - base
a - base is under attack
c - enemy unit is checked
m - achieving e->b->f->c by moving friendly unit closer to base

we have radix tree
ebfc
eba
am

note that we can feedback 'a' back into the radix tree to get action 'm'

of course some permutations, normalization or sampling of input is required, so that both words 'eb' and 'be' get equivalent state.

>>5
The animal brain has a hierarchy of specialized parts. It doesn't use the same mechanism at every level.

>>8

Neocortex does use the same mechanism. And this thread is not about simulating animal brain, but about exploring the possibility of dimensional radix trees.

>>9
Neocortex's a jew

>>10

is he circumcised?

>>11
Yes
I know that because I was sucking his dick yesterday

Markov chains only have a finite context. Your AI, even if it solves a dozen other Hard problems with its absurd design, is going to be unable to handle the question

What is 1+1+1+1+1+1+1+1+...+1?

for some large number of +s, unless you add some fundamentally new mechanism to it that goes beyond the scope of Markov chains (at which point, you should name your AI after the new mechanism instead).

>>13

for some large number of +s,

video game maps typically have 128x128 dimensions, so +s is fairly limited.

>>14
There is only one poster who could be so oblivious. Begone, Brazil-kun.

>>9
No, it doesn't. There are a myriad of models just for the vision processing part alone. From Self-Organizing Maps to Hierarchical RNNs. The brain has hierarchies of very specialized filters for input spaces to convert them into efficient and consistent representations. Kindly fuck off and inform yourself before spouting bullshit.
>>15
I was getting that feeling, as well.

>>16
AFAIK from an introductory anatomy lectures, brain does simple edge detection filter, then breaks edges into features, and then does pattern matching,using some advanced form of markov chains. With hearing it is even simpler: ear does FFT, and brain just pattern-matches resulting formants, which you can simplify to latin letters.

http://en.wikipedia.org/wiki/Feature_detection_%28nervous_system%29

>>17
If it were so simple. computer vision wouldn't be such an underdeveloped field, fuckwit.

>>18
please stop. you're name calling is harming the atmosphere of /prog/

>>18

Computer vision is a well developed field. For example, Microsoft's Kinect produces high quality 3d scans. The problem is that you mistake computer vision with the so called "Strong AI" chimera, which successfully avoided definition for the last 60+ years (and don't stat the "Turing Test" nonsense). That is where all these crackpot Jews, like Kurzweil and Minsky, come to the scene. I'm so happy, they went out of business, when AI winter came and goyim decided to cut funding the goal-less research.

>>20
Or the Vita that has an AR function to let you take photos of your waifu in natural settings, like your basement.

>>20

For example, Microsoft's Kinect produces [b]high quality[/b] 3d scans

Not really.

>>22

http://en.wikipedia.org/wiki/Kinect#Kinect_for_Xbox_One_2

Kinect for Xbox One ... has been praised for its wide angle, its fast response time and high quality camera. However, the Kinect's inability to understand some accents in English was criticized.[146] Furthermore, controversies surround Microsoft's intentional tying of the sensor with the Xbox One console despite the initial requirements for the sensor being plugged in at all times having been revised since its initial announcement. There have also been a number of concerns regarding privacy.