What is deep learning?
There are so many buzzwords in the field these days: big data, machine learning, deep learning, artificial intelligence, etc. How are they all related? In his book Deep Learning with Python (DLP), Chollet uses a Venn diagram to clarify the relationship between the last three. I added big data to his diagram:
Therefore, deep learning (DL) is simply one of the subfields in machine learning (ML).
The central problem in ML is to transform input data into meaningful output. In other words, to learn useful representations of input data that help us get to the expected output. This defines the meaning of learning. DL is one approach to learning. It learns using successive layers of gradually more meaningful representations. Since these layers are built on top of each other, if we try to visualize them in a graph, this graph can be deep. Hence, the origin of the name deep learning. The number of layers defines the depth of a model.
The hardware requirements to implement DL models are higher than the classical machine learning I started with. Running DL models on CPU (central processing unit) can be excruciatingly slow if it’s even feasible to do so. There are 2 solutions:
- get a GPU (graphic processing unit) from NVIDIA, currently the only one supports DL
- rent a GPU from places such as Amazon and run DL models on the cloud
Fortunately, the computer I have has a GPU, so I decided to use it for the time being and look into the cloud-based solution at a later date.