deep learning is the hottest thing (fad?) in AI. i know that WL has implemented a function, ImageIdentify, which i assume to be based on use of a convolutional neural network , but i was wondering, has anyone written a program for a CNN itself in WL? It's nice to be able to use ImageIdentify but if we had a CNN program, we could apply it to the other things besides image identification. btw - a nice tutorial on deep learning is available here: http://www.datarobot.com/blog/a-primer-on-deep-learning/
I'm busy working on a full neural network framework, that will have both CPU + GPU training functionality. It will also have feedforward neural networks (like convolutional neural nets) + recurrent neural networks (LSTM's etc. that can be used for sequence modelling, like text).
Hopefully this will be done for 10.3.