Managing Uncertainty in Expert Systems (Kluwer International Series in Engineering and Computer Science, 143)

Rating:
Summary: An introduction to Expert systems and methods
Review: I actually took the course that this book is the textbook for, with Dr. Grzymala-busse as my professor, at the University of Kansas. I had studied with him a year earlier on a research project in rules-based programming, when he was putting together the notes for it, so I feel quite fortunate in having seen it coming together.

The book itself is an extremely lucid, point-by-point introduction to the logical methods used in deriving and interpreting the rules and data in rule-based expert systems.

It covers classification, elimination of redundant data and outliers, ruleset reductions, systems that manage probability ranges with fuzzy and bayesian logic, data collection from human experts (by interview or observation), and learning and teaching systems. These are the foundations of Knowledge Engineering, one of the branches of Artificial Intelligence.

All in all, this material is necessary background for anyone who is going to be implementing a rules-based (or partially rules-based) system of almost any kind. It will, step-by-step, teach you what the design choices are and how to use various different systems most effectively.

Dr. Grzymala-Busse is very consistent with his approach of first presenting the problem, then the theory, then several approaches, then at least one fully-formed algorithm for a solution. Most of the chapters contain pseudocode, to make the solutions presented absolutely clear.

He also reviews the approaches taken by real-world workers and systems and places each in context with its problem and its theory. All in all, it's an excellent book.

The reason it gets four stars instead of five is because it's not a complete reference for a modern KE; there are a lot of techniques outside of the rules-based methodology that a modern KE needs to be familiar with, and while this book is an authoritative reference on handling uncertainty in rules-based systems, it is relatively silent on systems which are non rules-based. As Dr. Busse said at the time, "systems which seem to work, but where we cannot say with certainty why, or systems about which little can be proven, will be excellent and interesting subjects for other books."

Rating:
Summary: An introduction to Expert systems and methods
Review: I actually took the course that this book is the textbook for, with Dr. Grzymala-busse as my professor, at the University of Kansas. I had studied with him a year earlier on a research project in rules-based programming, when he was putting together the notes for it, so I feel quite fortunate in having seen it coming together.

The book itself is an extremely lucid, point-by-point introduction to the logical methods used in deriving and interpreting the rules and data in rule-based expert systems.

It covers classification, elimination of redundant data and outliers, ruleset reductions, systems that manage probability ranges with fuzzy and bayesian logic, data collection from human experts (by interview or observation), and learning and teaching systems. These are the foundations of Knowledge Engineering, one of the branches of Artificial Intelligence.

All in all, this material is necessary background for anyone who is going to be implementing a rules-based (or partially rules-based) system of almost any kind. It will, step-by-step, teach you what the design choices are and how to use various different systems most effectively.

Dr. Grzymala-Busse is very consistent with his approach of first presenting the problem, then the theory, then several approaches, then at least one fully-formed algorithm for a solution. Most of the chapters contain pseudocode, to make the solutions presented absolutely clear.

He also reviews the approaches taken by real-world workers and systems and places each in context with its problem and its theory. All in all, it's an excellent book.

The reason it gets four stars instead of five is because it's not a complete reference for a modern KE; there are a lot of techniques outside of the rules-based methodology that a modern KE needs to be familiar with, and while this book is an authoritative reference on handling uncertainty in rules-based systems, it is relatively silent on systems which are non rules-based. As Dr. Busse said at the time, "systems which seem to work, but where we cannot say with certainty why, or systems about which little can be proven, will be excellent and interesting subjects for other books."