INFORMATION SCIENCES 36,157-177 (1985) 157 zyxwvutsrqp An Expert System Oriented to Complex Pattern Recognition Problems A. GIORDANA and L. SAITTA Dipartimento dt Informatica, Universitb di Torino, Via Valperga Caluso 37, IOI25-Tonno, Italy ABSTRACT The general framework zyxwvutsrqponmlkjihgfedcbaZYXWVUTSRQPONMLKJIHGF BIMBO, designed for acquiring knowledge on expert systems oriented to pattern analysis, is presented. zyxwvutsrqponmlkjihgfedcbaZYXWVUTSRQPON BIMBO, which is itself an expert system, works in a largely automated mode and makes use of a large body of general knowledge and of powerful heuristics. The learning methodology explicitly addresses the problem of the presence of errors and noise in the data and utilizes an approximate reasoning scheme based on fuzzy sets. BIMBO has been applied in a test cam taken from the field of continuous speech recognition, with very good results in increasing the system accuracy and decreasing the demands upon human time. 1. INTRODUCTION Traditionally, complicated problems of pattern recognition, such as continu- ous speech understanding, waveform analysis, and scene interpretation, have relied upon procedural approaches using well-established statistical or pattern matching techniques. This is especially true with regard to the low level analysis of the patterns, as, for instance, segmentation and primitive extraction. An exception is the field of medical consultation problems, whose solution has very early served as a test case for knowledge based methodologies. However, these new methodologies seem now to be ready, after a decade of experimentation, to cope also with the sort of applications previously men- tioned, in which the great amount of computation to be performed poses stringent requirements on the systems devoted to their solution. Several at- tempts to apply artificial intelligence techniques to speech processing [ll-141, waveform analysis [15, 161 and vision [l-lo] have recently appeared in the literature. In fact, knowledge based systems are supposed to acquire humanlike exper- tise in a given field through an understanding of the fundamental laws and the causal relationships hidden in the phenomena under consideration; in this way @Xlsevier Science Publishing Co., Inc. 1985 52 Vanderbilt Ave., New York, NY 10017 0020-0255/85/$03.30