No, it's down with genetic programming. Evolution is THE algorithm for finding good solutions in a vast solution space. But restricted to areas, where a fitness function can be formulated. Look it up. Playing with genetic programming techniques is real fun.
The same principles are used for evolving AI brains, for example. The best AI brains out there are evolved.
@smarthandsomeguy Actually, genetic algorithms are a fairly niche field. They work well sometimes, but it's easily hindered by local optima and number of iterations required.
The best AI systems are generally those that are tailored to a very specific field (whatever that might be; robotic sports, playing Go, chatbots, etc.) using specific models/algorithms rather than genetic algorithms.
Still, Eureqa is a very interesting piece of software.
That is not really true. Easily hindered until you introduce crossover. Sex is good! Local maxima are a problem for simple hill climbing algorithms though.
>"The best AI systems are generally those that are tailored to a very specific field"
Of course. But watch?v=r6F3lT1v79I
Those AI brains are evolved. Have fun trying to get such results without genetic programming. But I am biased. If you have a hammer, the world is full of nails. ;-)
I'm trying to let eureqa fit a rational equation ( ax/(b+x)+c ) and it seems to be having quite some trouble discovering it. Should I let it run overnight?
The ordering of solutions by complexity is ... ingenious. I wonder how efficient this will be modeling complex models to some really noisy data (log-periodic exponential growth for stock market crashes).
i'm going to use it on behavioural data from 1300 ppl where there are patterns that standard psychological analyses obscure.. hoping to get this done at Easter. i'll tell u how i get on if u want. Goodbye SPSS :)
I am really excited by the possibilities this algorithm offers. I thought it rather interesting that the program modeled an exponentially decreasing phase term. I have not seen anything like that before.
Is there any reason to think this could not be applied to a GPU to make the searches even faster?
It looks like it could greatly benefit from GPU support, as the limiting factor in GPGPU scenarios is memory access most of the time, which shouldn't be crucial for an application like this.
@smarthandsomeguy They key there is "evolution in the broadest sense." If you simply generalized evolution to "gradual improvement relative to a specific measurement," then you could claim all kinds of things about it.
Evolution on the scale of DNA and species is barely moving, by modern standards. Evolution on the scale of technology is accelerating. If you want to generalize that much, evolution is an exponential curve of progress that started with the big bang.
Amazing video!
TheeImmortalPhoenix 2 weeks ago
How about releasing the source code for this program ?
labobo 2 months ago 2
This has been flagged as spam show
@labobo >"How about releasing the source code for this program"
It's done with genetic programming. If you are a software developer, you can dig into the topic within a day or so.
smarthandsomeguy 1 month ago
Could it be that this is a simple star search of combinations of functions?
An almost brute strength technique abusing the speed of modern computers...
tansor36 11 months ago
@tansor36 >"this is a simple star search"
No, it's down with genetic programming. Evolution is THE algorithm for finding good solutions in a vast solution space. But restricted to areas, where a fitness function can be formulated. Look it up. Playing with genetic programming techniques is real fun.
The same principles are used for evolving AI brains, for example. The best AI brains out there are evolved.
smarthandsomeguy 1 month ago
@smarthandsomeguy Actually, genetic algorithms are a fairly niche field. They work well sometimes, but it's easily hindered by local optima and number of iterations required.
The best AI systems are generally those that are tailored to a very specific field (whatever that might be; robotic sports, playing Go, chatbots, etc.) using specific models/algorithms rather than genetic algorithms.
Still, Eureqa is a very interesting piece of software.
GeekProdigyGuy 3 weeks ago
This has been flagged as spam show
@GeekProdigyGuy >"easily hindered by local optima"
That is not really true. Easily hindered until you introduce crossover. Sex is good! Local maxima are a problem for simple hill climbing algorithms though.
>"The best AI systems are generally those that are tailored to a very specific field"
Of course. But watch?v=r6F3lT1v79I
Those AI brains are evolved. Have fun trying to get such results without genetic programming. But I am biased. If you have a hammer, the world is full of nails. ;-)
smarthandsomeguy 3 weeks ago
update please download link to originallink /eureqa_download
m8kzardoz 2 years ago
This looks like it's a user interface for a genetic algorithm. Is it?
somecomputergeek 2 years ago
This has been flagged as spam show
@somecomputergeek >"This looks like it's a user interface for a genetic algorithm. Is it?"
It is.
smarthandsomeguy 1 month ago
I wonder how this could be made to work categorical data? How would one define the function and fitness metric for y as probabilities?
bhactuary 2 years ago 2
Very cool. After an overnight form it found a form of the equation that's equivalent to what I had in mind.
jimhsu77479 2 years ago
I'm trying to let eureqa fit a rational equation ( ax/(b+x)+c ) and it seems to be having quite some trouble discovering it. Should I let it run overnight?
jimhsu77479 2 years ago
The ordering of solutions by complexity is ... ingenious. I wonder how efficient this will be modeling complex models to some really noisy data (log-periodic exponential growth for stock market crashes).
jimhsu77479 2 years ago
Has anyone tried using Eureqa to tease meaning out of large sets of demographic or polling data? Any other Social Science applications?
jtjohnson555 2 years ago 7
@jtjohnson555
i'm going to use it on behavioural data from 1300 ppl where there are patterns that standard psychological analyses obscure.. hoping to get this done at Easter. i'll tell u how i get on if u want. Goodbye SPSS :)
DarkLordOfThDoomZone 11 months ago
Great Program! Thanks!
jnrolf 2 years ago
Very impressive.
Vire70 2 years ago
I am really excited by the possibilities this algorithm offers. I thought it rather interesting that the program modeled an exponentially decreasing phase term. I have not seen anything like that before.
Is there any reason to think this could not be applied to a GPU to make the searches even faster?
googacct 2 years ago 10
It looks like it could greatly benefit from GPU support, as the limiting factor in GPGPU scenarios is memory access most of the time, which shouldn't be crucial for an application like this.
JXero4HawkF 2 years ago 3
@googacct >"I am really excited by the possibilities this algorithm offers"
Well - "that algorithm" (aka evolutionary principle) created the entirety of currently observed biodiversity.
When you look at chimps using tools, you might think: "Wow, amazing what evolution can produce." But you have to say the same about human tools.
Hubble is a product of evolution in the broadest sense.
Here it comes: Eurequa is a product of evolution itself!
Think about that for a minute. ;-)
smarthandsomeguy 1 month ago
@smarthandsomeguy They key there is "evolution in the broadest sense." If you simply generalized evolution to "gradual improvement relative to a specific measurement," then you could claim all kinds of things about it.
Evolution on the scale of DNA and species is barely moving, by modern standards. Evolution on the scale of technology is accelerating. If you want to generalize that much, evolution is an exponential curve of progress that started with the big bang.
GeekProdigyGuy 3 weeks ago