Computers can learn from experience

International Conference on Case-Based Reasoning meets in Seattle, July 20-23, 2009

Suppose you want to make your world-famous, green-tomato salsa for your friends. Experience tells you that your friends enjoy eating salsa and chips, and that your recipe is a crowd pleaser. But you want to experiment and add black-eyed peas to the recipe. And you need to make a double batch because you have so many friends. In figuring out how to make the salsa recipe fit your present needs, you are using "case-based reasoning" — calling on past experiences and adapting them to find a new solution.

What if you could write a computer program that does that for you? A team of University of Washington Tacoma students is stirring up just such a program, under the guidance of Isabelle Bichindaritz, assistant professor at the university's Institute of Technology. The students are competing in a Computer Cooking Contest, one of the highlights of the eighth biennial International Conference on Case-Based Reasoning, the leading conference on this methodology.

The 2009 conference convenes next week, July 20 to 23, at the University of Washington campus in Seattle. Bichindaritz, who teaches artificial intelligence, data mining and bioinformatics at UW Tacoma, is the local conference chair. About 100 prominent computer scientists from all over the world will gather to share ideas and explore new ways to solve problems.

In the Computer Cooking Contest teams compete to see who can write the best case-based-reasoning computer program able to devise recipes and menus. Contestants are given basic recipes and a list of ingredients. Their computer programs must accomplish a set of tasks, such as selecting ingredients and inventing new recipes to fit a given set of parameters. The Computer Cooking Contest provides a concrete example of how case-based reasoning works.

In its computer cooking debut, the UW Tacoma team faces more seasoned teams that competed at the last conference. The UW Tacoma team is the only one from the United States.

In case-based reasoning, Bichindaritz said, computer programs actually learn from experience.

"These programs learn by interacting with their users and capitalize on past experiences to propose solutions to new problems, reasoning by analogy," she explained. Her research focuses on the applications of case-based reasoning in medical and biological fields.

She gives an example of a practical application. In a medical facility, a computer could be used to help diagnose a patient's case by "remembering" similar cases from the past. The computer program compares previous cases and adapts the solution to the new patient, taking into account differences in the patient's age, size, gender, medical history and other characteristics.

Bichindaritz likens case-based reasoning to the role memory plays in human problem-solving. Memory and intelligence are believed to be tightly linked in the process of human thought.

"In the first decades of computational intelligence, researchers thought that to design computer systems that would reason like humans, they needed to focus only on the inferential aspects of reasoning. Later, they realized that memory plays an essential role," she said.

Case-based reasoning studies how to write computer programs capable of learning from experience and getting smarter through interaction with human users. The method complements other computational intelligence methods, which are more focused on inferential aspects, Bichindaritz explained.

Although "case-based reasoning" is not yet a household phrase, Bichindaritz notes the field is well known in the academic areas of data mining and computational intelligence.

"Humans use analogical reasoning all the time to accomplish daily tasks," Bichindaritz said, adding it is often used for medical diagnoses, software design or architectural design. "But what about computational intelligence in the kitchen? That is less well known, but it opens a door to many variations in our day-to-day life."

The conference is sponsored by the U. S. Defense Advanced Research Program Agency, the U.S. Naval Research Laboratory, The Boeing Company, Verdande Technology in Norway and Empolis in Germany.

For more information on the Computer Cooking Contest and the International Conference on Case-Based Reasoning, visit www.iccbr.org/iccbr2009 or call Bichindaritz at 206-695-2080.