Deep Learning is an overrated computational technique due to recent hype in AI. Concerning the market enthusiasm, however, the course starts from the mathematical fact that deep learning is no more than a computational model for finding non-linear function.

Compared to SVM, deep learning powered by back-propagation + feed forward model is an inferior model in a sense that initial starting point matters in great deal, structure by itself is prone to vanishing gradient problem, and the model is unable to self-correct unlearn cases by outlier data points, as is for Decision Tree, one of the base model for Deep Learning (in the context of back-propagation + feed forward).

Once students understand limitations of the computational model, the course journey touches Auto-encoder in a sense of non-linear Factor Analysis, CNN for image recognition, and RNN for serial data processing. The course will conclude with a brief introduction of Boltzmann Machine, which will be the central topic of Advanced Deep Learning.