Similarly to the previous year, this workshop was hands-on : After a brief background on deep learning, participants started quickly, getting an understanding of features (and feature creation) by a ‘blame assignment’ algorithm (aka Backprop).
For this FOSSASIA workshop, I also created a stand-alone CNN filter demo, so that people could get an intuitive understanding of what a single 3x3 CNN filter can do.
Then, using the VirtualBox VM-on-a-stick, we had a quick look at a working GoogLeNet model. And then went on to a longer segment on how the CNN mechanism can be applied to Speech Recognition - a new application/module I put together specifically for FOSSASIA (in the preceeding 5 days, from concept to working & demonstrable code).
Naturally, this being a FOSS event, all the source is available on GitHub - if you have questions on the software, please leave an ‘issue’ there.
If there are any questions about the presentation please ask below, or contact me using the details given on the slides themselves.
The presentation was kindly recorded by Engineers.sg.
blog comments powered by Disqus