3 ms·
> Backpropagation suffers from vanishing gradients on very deep neural nets. Especially when using saturating functions (tanh/sigmoid) > Or the weights could
by raverbashing 12y ago
> Backpropagation suffers from vanishing gradients on very deep neural nets.
Especially when using saturating functions (tanh/sigmoid)
> Or the weights could be evolved using Genetic Programming
GA, not GP http://en.wikipedia.org/wiki/Genetic_algorithm http://en.wikipedia.org/wiki/Genetic_algorithm