This is lesson 3 of Machine Learning 101.
We are going to examine classes of machine learning problems.
In Lecture 2 we already mentioned a handwriting recognition problem. This is an example of supervised learning problem.
1. Supervised Learning
In supervised learning problems, you are given a set of data (training data) which is made up of set of input vectors. Then you are also given a set of outputs, or target vectors (that is the actual numbers).
Now, supervised learning is divided into two categories: classification and regression
In a case such as the handwritten digit recognition, where the objective is to assign each input vector to on of a finite set of discreet categories (numbers from 0 to 9), then this is called classification.
However, if the output is made up of a number of continuous variables (heights of people), then this is a regression problem. An example of regression is to predict the volume of sales for a marketing firm if certain amount was spend on an ad campaign in TV and newspaper.
So now, let’s summarize what you’ve learnt:
- Machine learning is divided into two categories: supervised learning and unsupervised learning.
- Supervised learning is divided into two: classification and regression.
Let’s now talk about unsupervised learning
2. Unsupervised Learning
In case of unsupervised learning, you have a training data set made up of a set of input vectors x, without and target values. The objective of unsupervised learning is to figure out trends in a given data set. In this case, it is called clustering. In clustering you need to find clusters or groups in the dataset that are similar.
If however, the goal is to determine how the dataset is distributed withing the input space, then it is called density estimation.
Then it could also be to reduce data from a high-dimensional space to a lower dimensional space. In this case it is called dimensionality reduction.
Let’s now summarize what you’ve learnt so far:
Unsupervised learning is divided into three categories: clustering, density estimation and dimensionality reduction
3. Reinforcement Learning
This is a relatively new area compared to the previous two. In reinforcement learning, the goal is to find a the best action to take in a specific situation so as to maximize some reward. In this case, no examples of optimal output is provided. The algorithm has to figure it out by a process of trial and error.
This is applied in game playing where the game have to figure out the best move to make to get it closer to winning.
However, we would not border much about this particular category. Just know that credit assignment problems falls under this class.
Now, the summary of all the classes of machine learning problems are given in the figure below: