Anytime I am talking to people about Machine learning and Artificial Intelligence and I mention “classifier systems”, there are always some people who are surprisingly ask, “what”? If you are one of those people who have never heard about classifier system as a method of learning, you should probably need to read couple of my upcoming posts.
Learning classifier system (LCS) is the first generation of classifier systems that was introduced by J. Holland in 1976. LCS is rule-based machine learning method that combines discovery and learning components and it founds its application in robotics, data mining, and data-driven prediction. From 1990, many different classifier system models (Like XCS, TCS, and ZCS ) were presented to solve real-word problems. Explained by John Holland, classifiers systems are:
- Condition/ action (If-Then) rules, simply a broadcast language
- Using genetic algorithm to learn
- Viewing environment by convey it to the set of detectors
- Acting upon an environment by detectors as well (packet of information, messages)
- Using messages for internal processing
- Learning through measure of their performance on the environment (feedback or payoff)
Moreover, there are three important factors regarding the mechanism of classifier systems;
“Parallelism and coordination are addressed by restricting rule action to the emission of messages.” In other words, set of rules are employed to act upon an environment based on the specific current state of the environment.
“Deciding which rules in a rule-based system are responsible for its successes, particularly when long sequences of ‘stage-setting’ actions precede success, is an interesting and difficult problem” which is defined as Credit assignment. In classifier systems, each rule bids to become active and then stands to profit from bids of subsequent bidders. In classifier systems, credit is accumulated by the rule as strength.
This was a very short introduction for classifiers systems. Please check my next blog posts if you want to know more. I will give you a short summary of the book “Learning Classifier Systems“.