Case-based reasoning (CBR) is a problem solving paradigm that is different from other major A.I. approaches. A CBR system can be used in risk monitoring, financial markets, defense, and marketing just to name a few. CBR learns from past experiences to solve new problems. Rather than relying on a domain expert to write the rules or make associations along generalized relationships between problem descriptors and conclusions, a CBR system learns from previous experience in the same way a physician learns from his patients.
The ability to retrieve and manipulate past problem-solving examples accurately is important problems in many areas, including:
- Process Control
- Advanced Manufacturing
The goal of Case-Based Reasoning
Case-based systems search their case memory for an existing case that matches the input specification. As new cases are solved they are added and continue to add to the database of cases solved.
The goal is to find a case that matches the input problem and goes directly to the solution, making it possible to provide solutions to potentially complex problems quickly. If an exact solution does not exist, the system will find one similar. The system learns over time, so that when a imperfect match is found that still solves the problem, the case is added to the system’s case memory for future use.
Learning is a key part of a CBR system’s architecture.
To understand how CBR works, one must understand what a case is to a computer. In the simplest form, a case is a list of features that lead to a particular outcome. Some examples on a credit application would be: credit cards, amount of loans outstanding, value of assets, etc. In it’s most complex form, a case is a connected set of sub-cases that form the problem-solving task’s structure — for example, the computer chip on a computer motherboard. The designs of the computer chip and the computer are made up of sub-designs, each of which could be considered a case unto itself.
One of the key differences between rule-based and case-based knowledge engineering is that automatic case-indexing techniques drastically reduce the need to extract and structure specific rule-like knowledge from an expert -– the most time-consuming part of rule-based knowledge engineering.
CBR systems derive their power from their ability to retrieve relevant cases quickly and accurately. Figuring out when a case should be selected for retrieval in similar future situations is the goal of this case-indexing process.
Building a structure or process that will return the most appropriate case is the goal of the case memory and retrieval process. Case indexing processes usually fall into one of three kinds: nearest neighbor, inductive, or knowledge-guided. Combinatorial approaches also exist.
Learning and generalization are critical to CBR systems.
Taking advantage of existing techniques for extracting useful information from examples lets case-based systems avoid some of the main problems of rule-based approaches in gathering problem-solving or classification knowledge and putting it to good use.
As cases accumulate, case generalization can be used to define prototypical cases that can be stored with the specific cases, continually improving system accuracy over time.
The inductive-indexing capabilities in CBR systems provide several major advantages over pattern-recognition techniques. Inductive systems can represent and learn from a wider range of feature types. The ability to use richer feature sets for describing examples makes them at least as accurate and many times more precise.
Limits of CBR
There are limits to CBR technology. The most important limitations relate to how cases are efficiently represented, how indexes are created, and how individual cases are generalized.