4 17.1. What Is Learning ? The machine cannot be called intelligent until they are able to learn to do new thing and adapt to new situations, rather than simply doing as they are told to do. “... changes in the system that are adaptive in the sense that they enable the system to do the same task or tasks drawn from the same population more efficiently and more effectively the next time.“ [Simon, 1983]
5 Rote-Learning Techniques Skill refinement vs knowledge acquisition Rote learning Taking advice Learning through problem solving Learning from examples
6 17.2. Rote Learning Rote learning is basic learning activity Caching has been used in AI programs to produce some surprising It exploited two kinds of learning : - rote learning, which we look at now, and parameter (or coefficient) adjustment. - rote learning of this sort is very simple, involve sophisticated problem solving capabilities, include : organized storage of information and generalization.
9 17.4. Learning in Problem Solving Can a program get better without the aid of a teacher? It can, by generalizing from its own experiences! Learning by parameter adjustment Learning with macro-operators Learning by chunking The utility problem
10 Parameter Adjustment Many programs rely on an evaluation procedure that combines information from several sources into a single summary statistic How much weight should be attached to each component Credit assignment problem
11 Macro Operators and Chunking Sequences of actions that can be treated as a whole Avoid expensive recomputation A chunk is essentially a large production that does the work of an entire sequence of smaller ones Several chunks may encode a single macro operator, and one chunk may participate in a number of macro sequences.
12 Learning with Macro Operators Start : ON(E,C) ON(D,B ) D B E CA D B A CE Goal : ON(A,C) ON(D,B) [Korf 1985b] It turns out that the set of problems for which macro-operators are critical are exactly those problems with nonserializable subgoals. Nonserializability means that working on one subgoal will necessarily interfere with the previous solution to another subgoal,
13 Learning by Chunking Chunking is process similar in flavor to macro-operator. The idea of chunking comes from the psychological literature on memory and problem solving. Its computational basis is in production system. Chunking is a universal learning method. SOAR system solves problems by firing productions, which are store in long-term memory.
14 Utility Problem While new search control knowledge can be of great benefit in solving new problem efficiently, there can be also some drawbacks. PRODIGY maintains a utility measure for each control rule: Average savings provided by the rule Frequency of its application Cost of matching If a proposed rule has a negative utility, it is discarded