neural network - Clockwork RNN (CW-RNN) and TanH activation function -
thanks again taking time answer post !
quick question :
in clockwork reccurent neural network (rnn), (here documentation),
seems me tanh-activated output layer suffer extreme load many connections, , alway output near 1 , or mistaken ?
let's have cw-rnn with, hidden layer, 4 modules of 50 neurons each.
even if these modules have different clock rate, sometimes, more 1 module activate & output else 0 : thus, output layer may output 1 because of many inputs.
is there missing in cw-rnn concept ?
know weight initialization, i'm wondering if missing piece of puzzle.
Comments
Post a Comment