It seems to me that in this 5-course specialisation, the first three course deal with more traditional ANN with more layers (and other common topics in machine learning) and the remaining two deal with RNN and CNN.
I am trying to understand:
- During the recent hype of deep learning, which one of these three generates the most useful applications? Among academics, when they use the term deep learning loosely, which one do they usually refer to?
- If the traditional ANNs with many layers did demonstrate its success and relevance in the last decade or two, what are the improvements which enables this resurgence? Is it enhanced computational power which make it possible to calculate very large ANNs (still the same activation functions and learning algorithm as before) which was not possible 20 years ago? Is it the sheer size of ANN rendering older learning algorithms insufficient and so that newer learning algorithms were developed?
@yinshan Welcome to the specialization! I figured I would share some thoughts from my perspective, and hopefully others from around the community will jump in and offer their perspectives also. So to answers your questions - as your work through the specialization you will see how each of these architectures are applied to real-world problems.
Convolution Neural Networks (CNNs) are a prominent part of Deep Learning algorithms for Computer Vision, but also make an appearance in other areas such as natural language processing. Recurrent Neural Networks (RNNS) are featured in Natural Language processing and time series predictions.
Each course in the specialization builds on the knowledge from the previous courses. So the foundational knowledge you gain from Course 1, 2 and 3 apply to understanding CNNS and RNNs in courses 4 and 5.
Hope that helps.
Dear @yinshan a warm welcome in the specialization also from my side. What I have experienced in the specialization is that, while attending all the courses, all the valid questions you made around ANN, CNN and RNN are explored from the theoretical side and also with practical examples. You will also discover the latest advancement in different field and why CNN or RNN might be better suited for certain a kind of problem and, more than that, you will figure out the general concept around picking the right tool for the AI problem you have.