1. Introduction
If you have tried to understand ANN before, it may not have seemed very user-friendly to you. The large and complex model can indeed be difficult to grasp. However, regardless of that, ANN is still a machine learning model, containing all the essential elements of a machine model. So, what are the basic elements of a machine learning model? Next, let’s review what we mean by a machine learning model first.
Mathematicians are right. Essentially, any machine learning model can be understood as a transformation \(f\) that converts feature variables into predictions. The type of model is determined by \(f\), which can be a simple linear classifier or other complex nonlinear models, while its specific characteristics are controlled by parameters. In other words, a successful machine learning model consists of the right model class \(f\) plus the appropriate parameters \(w\).
In the first part of the course, we used cars as a metaphor when we discussed the hyper-parameters, and now I’d like to introduce another one: shoes. Shoes come in various types, and we choose different kinds based on our purposes. For example, if you’re going to the beach, you definitely wouldn’t bring a lot of high heels. In machine learning, we determine which model to use based on the problem type, variable types, and sample size. Once you’ve determined the type of shoe you need, you then select the specific model and size that fits you best. In this analogy, the type of shoe represents \(f\), while the specific shoes model is hyper-parameter that you have to determine it first, and the size is model parameter \(w\). To complete the analogy, the input \(X\) would be your foot, and the output \(y\) would be your comfort or experience.
Do you like this metaphor, well I’ll be using it again later. Now, we can embark on our journey toward understanding ANN.
