Machine Learning

Machine learning is about giving more autonomy to the computer. You can interpret this as computers finding patterns through a set of examples (Supervised Learning) or without them (Unsupervised Learning), and computers adapting itself in dynamic environments (Reinforcement Learning).

In Supervised Learning often the specialist/teacher is a person that tells how the examples are classified. There is a training set (examples), and a test set. The examples are a pair composed by an input associated with an expected output, like a form of labeling. The task of this type of machine learning is to learn a mapping (given by a discriminant function) that generalizes the examples so it can be used with new data. This generalization is important because the training examples may contain outliers (data that represent noise or exceptions), or they may not contain one sample of every possible class. 

When we can't see the patterns in the input, we design an algorithm to classify it. Since there is no human interference, in the Unsupervised Learning we can't measure the error or give a reward for the candidate solution. So, we can only estimate the parameters using statistics methods as expected value, covariance matrix, and tensors.

If you thought that it was too much of statistics basis, be prepared. Reinforcement Learning is completely formulated as a Markov Decision Process (MDP), which is a matrix representing the probability of transition of a state to another. This method of analyzing the possible best next state makes the RL quite similar to dynamic programming. Believe me, when I studied about RL, there was a lot of probability calculus.

Of course, this three categories are compatible with the kind of problem you are dealing. The applications for SL are:

• Object detection/localization
• Action recognition
• Speech transcription
• Optical Character Recognition (OCR)

For USL:

• Redundancy Reduction
• Data Mining
• Text Mining
• Clustering

For RL:

• Path planing
• Game controller
• Robot locomotion
• Robot soccer

References:

1. Difference between SL, USL, RL
2. Applications of Unsupervised Learning
3. Applications of RL