4 ms·
Genetic optimisers don't tend to do well over a few hundred parameters, but stochastic gradient descent does well up to billions of parameters as long as you ha
by londons_explore 8y ago
Genetic optimisers don't tend to do well over a few hundred parameters, but stochastic gradient descent does well up to billions of parameters as long as you have sufficient compute and ram. Obviously your fitness function needs to be differentiable with respect to the parameters, which most simulators can't yet do.