Neural Networks and Deep Learning - deeplearning.ai
I just finished Neural Networks and Deep Learning course (first part) created by deeplearning.ai on the Coursera. I would like to write a little bit about my experience with it and also provide a little review.
The course is divided into 4 parts (also called weeks). There are deadlines, but I took only a free version of lectures so I try to make it in time but I don’t have to.
The first week was the introduction. There was information about deep learning in general and also about supervised part of machine learning.
In the second week was explained logistic regression and how to use it to binary classify (is or is not) pictures of cats. There was also a very informative part about training (gradient descent and derivatives in general). For those which are not familiar with python, there was a section which explained how to vectorize the calculation and what is the broadcasting in python.
The Third week was all about the shallow neural network (only with “one” layer). The new activation function was introduced, it is called “relu” and comparision with (mainly advantages over) sigmoind (which we still be using on the output layer).
Forth week was, I think the most important in the whole course, because there is the explanation of deep neural network with focus on implementing it. I really like a video called “Getting your matrix dimensions right”, which I have watched several times and it is intuitive explanation how to check that your dimension in the neural network are matching together. It can save you a lot of time in the debug mode.
I would strongly recommend to take part in this course. It is lead by Andrew Ng, one of the leading scientist in machine learning today. He is also a good teacher. I really like these programming assignments in every week with a code that was prepared exactly to let you focus on the deep learning part and not on the boilerplate code.