As of late, Machine Learning is the new hot buzzword being blown across the entire internet galaxy. The real questions are, “What is it” and “how can I use it”?
Let’s start with the three types of learning styles machine learning algorithms offer
- Supervised Learning
- Unsupervised Learning
- Semi-Supervised Learning
Before we jump into it, let’s go over some definitions.
- Model – The thing that is learning and that will spit out answers.
- Training Data – Data that will be used by the model to learn from
- Algorithms – The approach used to solving a problem. Think: Muay Thai vs Judo. Some obstacles in life need a fierce elbow to the face. Somethings in life need to be wrestled into submission.
1. Supervised Learning
The Big Picture: Supervised Learning is a lot like telling your kids the difference between right and wrong and expecting figure things out based on what you’ve taught them. If they’re not getting things right, we probably haven’t shown them enough examples to derive the correct answer just yet.
The Nitty Gritty: The training data, (data used to teach the model), has the answers attached. These answers are called ‘labels’. Labels will indicate whether the a tumor is benign or cancerous, the value of a house, or even whether or not you should continue reading a blog post. Kidding, not kidding.
Usually when we know the right answer, problems include regression and classification.
Algorithms may include linear regression and logistic regression.
2. Unsupervised Learning
The Big Picture: This is a lot like waking up at a poker table. Except, nobody speaks English, the rules are totally different, and instead of the regular 52 cards, there’s pictures of Sailor Moon characters.
It sounds awesome, because it is. Especially if you end up winning.
The Nitty Gritty: The situation is that the input data has no known target result and the model has to find patterns and correlations in the data. Problems for this type of learning include clustering, dimensional reduction, and association rule learning.
The most popular algorithm for this type of learning is the KMeans (K-Means).
3. Semi Supervised Learning
The Big Picture: This is basically like the poker table example, except getting thrown hints every once in a while.
The Nitty Gritty: With a small amount of labeled data, the learning speed and accuracy of the model improves considerably.
Problems for Semi Supervised Learning also include regression and classification.
Yeah money isn’t a learning algorithm, but it sure as heck has a lot to do with it. The costs of human labeled data can be considerably costly, while completely unsupervised data is cheap. Semi-Supervised Learning can be an option to significantly reduce costs while getting the right result as quickly and accurately as possible.