machine learning - How is the desired output of a neural network represented so as to be compared with the actual output? -
machine learning - How is the desired output of a neural network represented so as to be compared with the actual output? -
a neural network learns approximate desired output , hence 1 can utilize unknown testing sets , classify each testing illustration according respective class. illustration neural network can larn classify handwritten digits , recognise handwritten '9' value 9.
how neural network compare output of training sample desired output? encoding desired output take in construction of neural network?
the main thought don't create 1 single output , inquire "what digit this??". create 1 output each digit, , inquire each 1 "is digit x??".
so, desired output must encoded 1xn vector, n number of classes. values 0, , value corresponding desired class 1. in case example, create 1x10 vector, , encode 0 as
[1 0 0 0 0 0 0 0 0 0]
one
[0 1 0 0 0 0 0 0 0 0]
and 9
[0 0 0 0 0 0 0 0 0 1]
for actual output, create perceptron 10 outputs. each output give value in [0,1]. output max value wins, , regarded 1, while others regarded 0. so, if output of network is
[0.1 0.05 0.02 0.92 0.4 ....]
these correspond votes digits [0 1 2 3 4 ..], number suggested classifier three
machine-learning neural-network
Comments
Post a Comment