Case-Based Reasoning: What Exactly Is It?

Why is this important? What is the point?

There are problems, situations and information requirements that occur so frequently that the exact solution is known. Equally, there are sometimes new situations that require a new solution and a lot of creativity. Between these two extremes lies the normal case, i.e. not completely new, but not exactly the usual either. These are the cases that tie up the most resources in day-to-day operations and for which CBR is the ideal support.

By mapping similarity, the computer is able to point out similar situations to the user. The computer enables the user to apply knowledge from previous incidents at the right moment, according to the motto "similarities are the prerequisite for understanding the world".

How did this idea come about?

The CBR model goes back to the work of Roger Schank and his students at the famous Yale University. Roger Schank is an American theorist of artificial intelligence and cognitive psychologist who worked on episodic memory in the field of cognitive science.

In their experiments, Schank and his colleague Janet Kolodner observed that experts, such as doctors or lawyers, did not store their knowledge and experience in the form of IF-THEN rules as expected, but in the form of specific episodes (case studies). When a new problem arises, they use precisely these stored episodes to solve the problem. They remembered similar cases from the past and simply transferred the successful problem solution to the current case. People usually solve problems by remembering a similar situation from the past and transferring the solution from that time to the current situation. Doctors, lawyers or service technicians: it is precisely this ability, combined with the knowledge of many examples from a long professional life, that makes up experience.

That was a completely new insight. If people were able to solve really complex problems quickly and reliably, this process should also be transferable to a computer. The two scientists called their approach CBR.

German CBR pioneers in Kaiserslautern

In the early summer of 1987, Klaus-Dieter Althoff, then a member of the DFG Collaborative Research Center 314 in Building 14 at the Expert Systems/DFKI Chair, came across these articles in the journal "KI - Künstliche Intelligenz" (AI - Artificial Intelligence) and was immediately hooked. If this theoretical approach from cognitive science could be implemented in practical computer programs, it would be a revolution in the technology of expert systems.

Together with several students and the mathematician and world-renowned logician, Professor Michael M. Richter, the idea was developed further and a new branch of research was born.

Over the course of many student research projects, projects and dissertations, doctorates, major European Union research projects such as INRECA I + II and in cooperation with many partners, including the Department of Mechanical Engineering at the TU, a very vague idea slowly developed into real software. This was successfully tested on specific industrial problems such as the diagnosis of CNC machine tools or the diagnosis of aircraft engines.

The first European "Case-based Reasoning Conference" was held near Kaiserslautern in 1993. In 1997, Empolis received the Rhineland-Palatinate Innovation Award for its CBR software, the first in a whole series of awards for the company, which is one of the CBR pioneers. CBR-Works was one of the first commercial CBR tools on the market. Empolis can therefore look back on more than 30 years of experience. A large number of applications in the most diverse application areas have been developed during this time.

Where is CBR used?

CBR can be used, for example, in the maintenance of machines and plant knowledge and service systems, without additional modeling effort. In addition to expert knowledge, CBR can be used to make decisions based on machine status data - including other sources of information such as ticket systems - without the need for human experience. This is of crucial importance, as the detection of complex incidents often depends on so many events that manual, rule-based modeling would be far too time-consuming, too difficult or even impossible. CBR enables the early detection of faults before major damage occurs. Such a system is able to quickly identify the cause of a fault by gradually narrowing down the source of the fault.

Case-based reasoning thus enables early fault detection and problem-solving based on similar incidents in the past. Ad hoc conclusions about the most probable cause are generated in order to successively narrow down the result set.