How Do Machines Learn?
What is this?
I'm reading Grokking Deep Learning and am going to put my notes here. This is from Chapter 2 - How Do Machines Learn?
What is Deep Learning?
Deep learning is a sub-field of Machine Learning that primarily use Artificial Neural Networks.
What is Machine Learning?
Machine Learning is a sub-field of computer science where computers learn to do things that they weren't explicitly programmed to do. Their main goal is to map a data set to some other useful data set.
What is Supervised Learning?
Supervised Learning methods transforms one dataset into another. They take what we already know and try to come up with what we want to know.
What is Unsupervised Learning?
Unsupervised Learring methods group your data. They take your data and try to come up with labels for clusters within the data.
What are Parametric and Non-Parametric Learning?
What is Parametric Learning?
- Parametric: trial-and-error (has a fixed number of parameters)
- Non-Parametric: counting and probability (has an infinite number of parameters)
The classifications Supervised and Unsupervised refers to the pattern that is being learned, while Parametric vs Non-Parametric is about the way what's learned is stored.
What is Supervised Parametric Learning?
Trial and error learning that tunes your model's knobs.
- Step One: Make a prediction using your data
- Step Two: Compare your predictions to the real answer
- Step Three: Change your model based on how you did - make it more or less sensitive to each of the parameters
What is Unsupervised Parametric Learning?
It's parametric, so it has knobs to twiddle when finding groups, but the knobs are used to tune the input data's likelihood of being in a group.
What is Non-Parametric Learnining?
These are counting-based methods - the number of parameters depends on the data. If you have a set of labels relating to an outcome, each label might be a parameter and your model would count how many times each label lead to the outcome you're watching.