FGP: A Virtual Machine for Acquiring Knowledge from Cases

Large case databases are numerous and packed with information. The largest of them are potentially rich sources of domain knowledge. The FGP machine is a software architecture that can make this knowledge explicit and bring it to bear on classification and prediction problems. The architecture provides much of the functionality of traditional expert systems without requiring the system builder to pre-process the database into rules, frames, or any other fixed abstraction. Implementations of the FGP machine use similarity-based reminding and the cases themselves to drive the inference engine. By having the system calculate and incorporate a measure of feature salience into its distance calculations, the FGP architecture smoothly copes with incomplete data and is particularly well-suited to weak-theory domains. We explain the model, describe a particular implementation of it, and present test-results for a classification task in three application areas. 1 Introduction The knowledge acquisition bottleneck is still one of Al's major problems. It has engendered commentary from senior AI practitioners for years, sparked the current revival of interest in machine learning, and is the primary motivation behind a multi-million dollar project to assemble a massive knowledge base (on the order of 10 8 facts) [Guha and Lenat, 1990]. In recent workshops, many researchers have come to the conclusion that the effective use of large databases is our best hope to break through this bottleneck [Friesen and Golshani, 1989]. Our work is in this vein, and is an attempt to devise a domain-independent software architecture and related algorithms to extract knowledge from a large database of unstructured cases, a term we define below. Cases are similar to feature-vector representations of data, but more general. They may be incomplete; individual cases need to have values for only an arbitrary subset of the universe of features. Cases are not assigned to fixed categories. Rather any feature in the feature-universe is a potential category; the program must be willing to direct its inference process at determining the probable value of any arbitrary feature for a particular case. We say probable because we make no assumption about consistency of cases. Two cases may have identical values along every dimension but one. We adopt this definition because these are the characteristics observed in cases drawn from real-world domains. The question we sought to address is whether the static feature information stored in cases provides enough material to do useful inferencing and learning. The approach we …