Machine Learning is a subset of Artificial Intelligence. Computer programs should predict an outcome without beeing explicitly programmed.
We talk about Supervised Learning whenever the machine gets training data to learn from. The machine is „supervised“ by the data. Types of Supervised Learning are Regression and Classification.
Regression: A function is determined, which best fits the training data.
Classification: Categorize given data set, based on training data.
When using Unsupervised Learning, the data is unlabeled and the machine tries to find structure in the data.
Unsupervised Learning is used in computer vision and audio processing.
Sometimes a system has to constantly come up with decisions over time. That is when Reinforcement Learning comes in handy. The system doesn’t have to come up with a solution all at one, instead it makes a sequence of decisions which get better over time.
One method – which is inspired by training animals – is the Reward System, in which the computer system receives positive or negative Feedback based on the decision taken. The learning algorithm then differentiates between „good“ and „bad“ behavior.
This and the coming blog posts are based on Andrew Ng’s Stanford lectures which are available on YouTube for free. The blog posts are primarily used as a way to summarize what I have learned in layman’s terms. Details will be updated over time.