Proposition: Visualization Tool for CNNs

[Samy Melaine]

Hello,

I'm an AI master student who spent the summer trying to find out an interesting idea for my memoir (master thesis).

I'm really interested in deep learning so I explored it a little further and found the interesting paper published by Zeiler and Fergus about Visualizing and Understanding Convolutional Networks (which he explains in this video https://www.youtube.com/watch?v=ghEmQSxT6tw). So I wanter to use this idea to implement Visualization Tool for CNNs using theano & pylearn.

I just found out that Jason Yosinski (et al.) realized it few months ago: http://yosinski.com/deepvis as a caffe ( http://caffe.berkeleyvision.org/ ) toolbox, so my goal would be to implement it for pylearn.

What do you guys think about the idea? is it worth it?
Any thought, remark or critic would be great.
Thanks,