Pattern recognition for Learning AI
Author
I came up with the idea a few days ago, and am sure it has been thought of before, but I can''t find anything about it. The idea is to create a fairly basic learning AI from a pattern recognition algorithm. I don''t see why everyone is trying to figure out how to tell an AI system how to recognize english, when the system is created to learn. Why not just let it figure it out. Same with speach recognition. Why spend years of research trying devise an algorithm to convert sound into text when the application using the text is meant to figure things out on its own?! I am sure there is a falicy in my argument, but I haven''t been able to see it and can''t get the thought out of my head. It has also interested me in pattern recognition. Anyone know where to find some starting points?
Thanx ahead!
I am no expert on this, of course, but I believe there are a number of practical problems to applying your theory. For instance, pattern recognition is fairly complex and although it is getting better, it is still not near as good as a human brain for recognizing letters and words quickly.
Another issue is the fact that just because the AI can recognize a pattern doesn''t mean it has a meaning associated with that pattern. How do you explain what "the" means to an AI?
Assuming you can solve these problems, how do you actually implement this idea? There are various learning algorithms, but none of them are particularly wonderful. Many PhD''s have spent many years trying to devise robots or software that can learn and evolve it''s own behaviors. I believe there was a professor in Australia a few years back that had designed a robot that would learn how to use it''s sensors and arm on it''s own very gradually by ''experimenting'' with itself. I''m not sure what the results of this were, but you might want to try looking it up. I''m sorry that I can''t remember the name of the researcher or the robot.
Another issue is the fact that just because the AI can recognize a pattern doesn''t mean it has a meaning associated with that pattern. How do you explain what "the" means to an AI?
Assuming you can solve these problems, how do you actually implement this idea? There are various learning algorithms, but none of them are particularly wonderful. Many PhD''s have spent many years trying to devise robots or software that can learn and evolve it''s own behaviors. I believe there was a professor in Australia a few years back that had designed a robot that would learn how to use it''s sensors and arm on it''s own very gradually by ''experimenting'' with itself. I''m not sure what the results of this were, but you might want to try looking it up. I''m sorry that I can''t remember the name of the researcher or the robot.
Author
Thanx for the fast reply. I was sort of suspecting the complexity of pattern recognition. The problem I see is the starting point. In EVERY AI system i''ve seen there is a motive, or goal for that system. If I was correct in the fact that the AI could figure everything out from nothing, how would you describe its motives. Also isn''t it not true that the way we as humans learn is by recognizing patterns made in time... And start as a baby with absolutely no knowledge? Somehow we start from nothing and endup with a lot more. I guess we start out with the motivation to survive. From there we recognize that when we cry, we get taken care of. whether that is some food, protection, etc. How is this not possible for a AI system to recognize. Computers I''m sure would also not take 7 years before it comprend addition.
P.S. Anyone know where I could find some info on pattern recognition algorithms?!?!?!
Thanx alot ahead
P.S. Anyone know where I could find some info on pattern recognition algorithms?!?!?!
Thanx alot ahead
Author
Anyone know how to come up with a function based on data it is given. A Linear function would be fairly easy...
y = mx + b ''in 2D
or
f(x) = mx + b ''how ever you like it
m = slope
b = where line intersepts y axis
and in any other amount of N dimentions you can define a line by N - 1 slopes, and N - 1 axis intersepts. My problem is coming up with a standard for a parabola for example, and or parabolas in more than two dimentions. If anyone has any function formats, could you please let this post know.
Thanx alot ahead
y = mx + b ''in 2D
or
f(x) = mx + b ''how ever you like it
m = slope
b = where line intersepts y axis
and in any other amount of N dimentions you can define a line by N - 1 slopes, and N - 1 axis intersepts. My problem is coming up with a standard for a parabola for example, and or parabolas in more than two dimentions. If anyone has any function formats, could you please let this post know.
Thanx alot ahead
Equation fitting is an irritating topic in mathematics. You''ll find a bunch of info about this in any college level statistics textbook. The methods you''ll find in there are mostly for polynomials with a known degree, though.
If you''re still thinking about sound waveforms, try reading a little about fourier analysis and/or transforms if you want a more general (and much more difficult!) solution. Any search engine should give you plenty of links to fourier papers. I''ve heard it bandied about that this may be the way human brains decode visual/audio data.
If you''re still thinking about sound waveforms, try reading a little about fourier analysis and/or transforms if you want a more general (and much more difficult!) solution. Any search engine should give you plenty of links to fourier papers. I''ve heard it bandied about that this may be the way human brains decode visual/audio data.
Author
CheeseGrater, Thanx for the info. Everyone is really helping and more than I was expecting... I was expecting nothing, anyways... All help gratefuly accepted!
Zach Dwiel - Tazzel3D
Zach Dwiel - Tazzel3D
This topic is closed to new replies.
Advertisement
Popular Topics
Advertisement
