Dear /sci/,
Suppose that I've created an algorithm that will optimize an arbitrary small sequence of binary data (up to 1kb). This binary data can represent anything you want, from computer instructions, to DNA, to physical objects. The algorithm will recommend a sequence of bits which you then evaluate with a simulation technique of your choosing, and then you tell the algorithm how "good" of a suggestion that piece of data was. Based on your ratings of the different suggested data sequences, the algorithm will suggest a "better" sequence of data.
What are some interesting problems in your field of study that could make use of this sort of optimization? I'm looking for intriguing applications where the algorithm will basically design an optimal solution to your problem.
Thanks,
Anon Y. Mous
>>8160383
It is called supervised learning, dipshit.
>>8160470
That's not even remotely what this is. This is a global black box optimization technique for binary data. I'm publishing a paper about it, but I want to present some more applications.
>>8160383
Find the best pick-up line.
>>8160475
>global black box optimization technique
Are you a wizard?
>>8160383
>The algorithm will recommend
If in the first step the algorithm is not comparing your 'arbitrary data' to anything then is this not just LITERALLY randomly generating 128 bytes?
>OP made a random number generator in python and now he thinks this is /sci/ material
Please prove me wrong OP.
>>8160383
>"good"
>"better"
>""""optimized"""
since initially the simulation technique doesn't matter so no general response mechanism can be implemented, how does one "optimize" a sequence of bits while still maintaing uncorrupted files (for the original decoder) .
>Suppose that I've created
>I'm publishing a paper
whew buddy I almost mistook you for a srs poster.
how can you publish a paper about an algorithm you didn't even (re)create.
>>8160487
Interesting, but it would take forever to evaluate all of those manually.
>>8160516
No, it's just an extension of existing optimization methods for scalars.
>>8160519
It builds a statistical model of the search space and will continue to reccomend better and better sequences of data as it refines that model.
>>8160521
I'm not quite sure what you're asking. The simulation technique doesn't matter in the sense that it's a black box method. It just needs you to give it a score for the sample points that it suggests.
Now that we've all gotten the autism out of our systems, can anyone suggest some interesting applications that they might want to see?
>>8160475
It is exactly what this is. Are you retarded?
>>8160383
art, particularly music
>>8160796
Supervised learning refers to classifiers, not optimizers. Would you call a genetic algorithm a form of supervised learning? No? Then shut the fuck up you little kid. My research fulfills a similar purpose to GA's, albeit in a much more effective and advanced way. It seems like whenever I try to share something interesting with the community here the autists can't stop bitching and whining.
This is why you can't have nice things.
>>8160798
That would also have to be evaluated manually by a person, which would take quite a while since it probably need at least a few hundred function evaluations before it could converge, depending on the length of the bit sequence. Ideally the fitness function should be evaluateable by a computer so it can be done rapidly.
What other techniques have you benchmarked it against? Bayesian estimation? Random search?
Have you tried using it for hyperparameter optimization in machine learning? That's a big area of interest.
>>8160881
So, you are a retard. All you are doing is making a guess, asking me for feedback, and then improving your guess based on my feedback. That, by definition, is supervised learning. I can reproduce what you are doing in 30 minutes. Amateur.
>>8160916
It's a distant relative of bayesian estimation, and preliminary tests so far have been significantly better than random search. Hyperpamater optimization is a largely solved problem and pretty much just deals with scalar value. This approach deals with arbitrary binary data, so that wouldn't be using the algorithm's full potential.
>>8160941
>I could build my own space shuttle
>all I have to do is glue some wings on a rocket engine, rite guise? XDD
Embarrassing.
>>8160951
Good luck with your science fair project. I hope you win.
>>8160963
Ok kiddo. I'll be looking for all of your journal articles too.
