Education

Objective:

The serve as a means to of this will not only enable you to read 80% of the papers, but also know how the authors implemented it. There is a level of intuition that we can obtain by coding up all the methods. The goal is not just comprehension, but it is application. There is a tendency for the beginner to want to just "use" technology. Unfortunately, deep learning algorithms aren't just plug and play, and using someone else's code intimately related to knowing when to use it. Or simply, do we want to believe whether or not something is possible or not? Do we waste time replicating research of

Approach:

Understanding cost functions. Do we want to minimize an error with respect to some criteria? Do we want to normalize or regularize to induce some kind of sparse solution? Do we want to optimize for maximum entropy? Or is it

Syllabus:

Week Number	General Approach	Specific Implementations
1, 2, and 3	Supervised Cost Functions	Inverse Problem Gradient Descent Backpropagation
4 and 5	Unsupervised Cost Functions	ICA and variants (e.g., RICA) Beamforming
6 and 7		Deep ICA GANs with ICA Cost
8 and 9	Recurrent Methods	CLDNNs (Sainath et al) LSTM+RICA Cost