2 ms·
Genetic algorithms have been around in CS for a long time, in lots of highly effective optimization problems, particularly in structural problems. They don't s
by dnautics 8y ago
Genetic algorithms have been around in CS for a long time, in lots of highly effective optimization problems, particularly in structural problems. They don't seem to be as good for AI/ML problems (though the jury is still out in hyperparameter type stuff). Gradient descent via backpropagation is quite effective here, and that is very much not like how the dandelion pod evolved.
- neuronic 8y agoMaybe that's as meta as it gets about evolution. The mechanisms of evolution itself are put under selection pressure once we put alternatives out there. Once we truly break the barrier between metal and meat, artificial processes may create real life in completely unforeseen ways. Gives a new ring to "asexual reproduction".