It seems to me that genetic algorithms would be an ideal way to train neural networks so that they come to have the right weights, since they are especially good at escaping local minima, and converging on a global optimum. However, from my rudimentary knowledge of neural nets so far, I haven't come across this method in any tutorial video. Are there inherent challenges that make GA's unsuitable for this purpose? Or is there a better way to accomplish the same thing?
$\begingroup$
$\endgroup$
2
-
1$\begingroup$ "since they are especially good at escaping local minima, and converging on a global optimum" -- what makes you say that? $\endgroup$– RaphaelCommented Aug 10, 2017 at 7:00
-
$\begingroup$ "genetic algorithms would be an ideal way to train neural networks" -- how exactly would you want to do that? $\endgroup$– RaphaelCommented Aug 10, 2017 at 7:01
Add a comment
|
1 Answer
$\begingroup$
$\endgroup$
1
There have been hundreds of papers published over the years on training neural networks with GAs. Here's a starting point Neuroevolution.
The basic problem though is that stochastic gradient descent works really well in vastly less computation time.
-
$\begingroup$ I suggest to remove that bracket from the link. $\endgroup$– rus9384Commented Aug 10, 2017 at 14:00