Supervised, Unsupervised, Semi-Supervised, and Reinforcement Learning

Within artificial intelligence and machine learning, there are several different subcategories…

With Supervised Learning, you feed The Machine lots of labeled input and output data to train the algorithms to get a specific outcome. Then you “supervise it” by giving it constant, corrective feedback when it’s wrong.

With Unsupervised Learning, you don’t already know the answer/outcome, so your data is not labeled or organized ahead of time. The Machine’s goal is to identify patterns in the data sets on its own; figure out what stands out, and then tell you something you don’t already know. Unsupervised learning excels at pattern and trend identification. It pinpoints patterns, similarities, differences, and outliers and comes up with correlations.

Semi-Supervised Learning uses a small amount of labeled data and oodles of unlabeled data. You see this a lot in Medical companies/categories, where doctors will label a sampling of different cancers/diseases they’d like to study. The Machine will use this sampling along with an entire library of millions of other MRIs, CT scans, X-rays, etc., to find patterns.

With Reinforcement Learning, The Machine is not given explicit goals except to maximize reward/value. You provide The Machine a task, and it learns from experience and finds solutions. It often improves results because the Machine has agency and can change its mind. Think Pavlov’s Dogs.

Have questions about these or other types of learning? A tip you’d like to share? Tweet @amyafrica or write info@eightbyeight.com.

A Down-and-Dirty Definition for Marketers. (Read more about these here.)