I'm reading Nielsen's book on neural networks. In the first chapter you construct a neural net to recognize handwritten digits by training the network on 60,000 samples. I wonder how would you build a neural net to recognize a person by his/her silhouette? Is there some special technique to make it possible?
$\begingroup$
$\endgroup$
Add a comment
|
1 Answer
$\begingroup$
$\endgroup$
2
For this purpose you have to have a data-set that can be interpreted by the human. I mean an expert should label the data samples without hesitating. Then you can make a typical neural network and train it. In your case you have 10 classes and all 60 thousand images are labeled by an expert. Consequently, you are able to train a network. Your task is supervised and you need labeled data-set.
answered Feb 4, 2018 at 8:52
-
1$\begingroup$ Another idea is to take profile images and convert them to silhouettes - no need for an expert in that case. $\endgroup$– ImranFeb 4, 2018 at 11:05
-
$\begingroup$ @Imran By expert, I meant there should be someone who can figure out the data that may increase the Bayes error. In the feature space there may be similar input patterns that have different labels but so much identical input patterns. In those cases the Bayes error may increase. But it was too detail, so I didn't refer to :) $\endgroup$ Feb 4, 2018 at 11:28