I am working on an optimized method for measuring the similarity between 2 signals, Is it possible to use a genetic algorithm for finding the correlation among time series?
3
-
1$\begingroup$ Just making sure: do you mean "genetic" or "generic" ? $\endgroup$– HilmarNov 22, 2021 at 14:32
-
$\begingroup$ Genetic algorithm $\endgroup$– BlobmouNov 22, 2021 at 14:32
-
1$\begingroup$ Thanks for clarifying . Can you add maybe some background? What research have you done so far and why do you think that genetic algorithms are a good fit for this type of problem? Also it would help if you define "similarity" some more. What are the classes of operations/transforms that you consider to be "similar" versus "dissimilar"? $\endgroup$– HilmarNov 22, 2021 at 14:41
Add a comment
|
1 Answer
$\begingroup$
$\endgroup$
A genetic algorithm can converge towards a correlation, I see no reason why correlation would be any different than any other function mapping two signals to a real number... in fact, correlation is nice in terms of partial derivatives, so that most methods of optimization would converge on it, if the metric you're using actual has a minimum reached by the function known as "correlation". In cases where the "classic" optimization methods work (simple gradient descent?) it's probably not a good idea to throw genetic algorithms at the problem.
Why you wouldn't just start by using the correlation right away instead of "developing" it in sequence of functions – nobody knows but you. Why use an optimization technique if you know the optimum!
answered Nov 22, 2021 at 14:49