This is part of my notes taken while studying machine learning. I’m learning as I go along, and may update these with corrections and additional info.
Types of Machine Learning
There are three main branches of machine learning:
- Supervised Learning
- Unsupervised Learning
- Reinforcement Learning
- In Supervised Learning, algorithms learn from studying labeled data.
- After this, SL algorithms are able to determine which label is given to new data based on observing patterns and associating them to new, unlabeled data.
Supervised learning has two sub-categories:
Classification models predict categorical outcomes. They can answer yes/no questions, such as: Should the email be categorized as spam, or not-spam? Is the patient sick, or not?
Classification is not limited to just 2 outcomes. Any number of categories can be predicted with most models, such as Which type of Iris is this flower? Which breed is this dog?
- Used to categorize data
- Can be any number of categories
- Is this email spam or not spam?
- Which breed of dog?
- Which variety of Iris?
Regression models predict numeric outcomes, such as home value, or a person’s height. They can answer the questions of the form: How much?
- Used to predict numeric outcomes
- How much does this house cost?
- How tall is this person?
- How many seconds do you expect someone to watch a particular video?
Unsupervised learning creates models for data when there are no pre-existing labels to train on, therefore it is considered unsupervised. Some examples include:
- Building a music recommendation system
- Grouping similar items together
- Reducing a dataset using in another ML model to a fewer number of useful features
Reinforcement learning is learning based on taking certain actions, and receiving awards for those actions. Examples include:
- Self-driving cars
- Game playing agents, or bots, such as AlphaGo
Deep Learning has beaten all other machine learning algorithms in its ability to predict.
- Deep Learning can be used for Supervised, Unsupervised, and Reinforcement learning
- The rise in popularity of DL coincides with the rise in belief that we should care less about how we make predictions, and instead care more about the accuracy of our predictions
- With DL, we rarely understand how or why our model makes certain predictions
Three Barriers to Using Deep Learning
- It requires a lot of data
- It requires a lot of computing power
- There is no understanding of how or why decisions are being made, due to the complexity and flexibility of the algorithm
Deep Learning Examples:
- Identify when a movie was created based on the video footage