My name is Grzegorz Gwardys and I am a PhD student in Division of Television, that is a part of Institute of Radioelectronics and Multimedia Technology at Warsaw University of Technology. My research interests revolve around Deep Learning in multimedia indexing, mainly music tracks. 2012-2015 I was a member of a group concentrating on Artificial Intelligence in Samsung Electronics Poland – I have an experience both in Natural Language Processing (patent application A method and system for building a language model ) and Computer Vision. After hours, I am a president of MuGEd Scientific Group.
Grzegorz Gwardys 
Hey great article on conv net! – https://grzegorzgwardys.wordpress.com/2016/04/22/8/
Can you help me with a question on it?
For #3 on the bottom of the article:
If I am making a neural network that looks like this: http://pubs.sciepub.com/ajme/2/7/9/image/fig2.png
The Classification neural net will have an output layer of 10 neurons, making the Cost function C = 1×10 matrix
How can I multiply C by sigmoid prime of zL like in your #3 equation?
Isn’t sigmoid prime of zL (for the feature learning conv net) a 1x1x300?
How do I get delta L (delta of last layer of feautre learning – conv neural net)?
Thanks!
LikeLike