Reference: Sullivan, J. RL3: An Approach to Incremental Rule Learning. June, 1988.
Abstract: Most current induction programs process their training sets all at once. In incremental learning, the induction program accepts a portion of the examples at a time and modifies previous results without starting all over again. There are several potential advantages of learning incrementally, including lessened memory and disk space requirements, faster performance, and the ability to handle concepts which are changing over time. This paper describes RL3, a descendent of RL which among many other improvements has the capability of learning incrementally.
Notes: 24 pages.