TY - JOUR
T1 - Automated Concept Acquisition in Noisy Environments
AU - Bergadano, Francesco
AU - Giordana, Attilio
AU - Saitta, Lorenza
PY - 1988/7
Y1 - 1988/7
N2 - This paper presents a system which performs automated concept acquisition from examples and has been especially designed to work in noisy environments. The learning methodology is aimed at the target problem of finding discriminant descriptions of a given set of concepts and uses both examples and counterexamples. The learned knowledge is expressed in the form of production rules, organized into separate clusters, linked together in a graph structure; the condition part of the rules, corresponding to descriptions of relevant aspects of the concepts, is expressed by means of a language based on first order logic, enriched with constructs suitable for handling uncertainty and vagueness and increasing readability for a human user. A continuous-valued semantics is associated to this language and each rule is affected by a certainty factor. Learning is considered as a cyclic process of knowledge extraction, validation, and refinement; the control of the cycle is left to the teacher. Knowledge extraction is guided by a topdown control strategy, through a process of specialization. The system also utilizes a technique of problem reduction to contain the computational complexity. Moreover, the search strategy is strongly focalized by means of task-oriented but domain-independent heuristics, in an attempt to emulate the learning mechanism of a human being, who has to find discrimination rules from a set of examples. Several criteria are proposed for evaluating the acquired knowledge; these criteria are used to guide the process of knowledge refinement. The methodology has been tested on a problem in the field of speech recognition and the experimental results obtained are reported and discussed.
AB - This paper presents a system which performs automated concept acquisition from examples and has been especially designed to work in noisy environments. The learning methodology is aimed at the target problem of finding discriminant descriptions of a given set of concepts and uses both examples and counterexamples. The learned knowledge is expressed in the form of production rules, organized into separate clusters, linked together in a graph structure; the condition part of the rules, corresponding to descriptions of relevant aspects of the concepts, is expressed by means of a language based on first order logic, enriched with constructs suitable for handling uncertainty and vagueness and increasing readability for a human user. A continuous-valued semantics is associated to this language and each rule is affected by a certainty factor. Learning is considered as a cyclic process of knowledge extraction, validation, and refinement; the control of the cycle is left to the teacher. Knowledge extraction is guided by a topdown control strategy, through a process of specialization. The system also utilizes a technique of problem reduction to contain the computational complexity. Moreover, the search strategy is strongly focalized by means of task-oriented but domain-independent heuristics, in an attempt to emulate the learning mechanism of a human being, who has to find discrimination rules from a set of examples. Several criteria are proposed for evaluating the acquired knowledge; these criteria are used to guide the process of knowledge refinement. The methodology has been tested on a problem in the field of speech recognition and the experimental results obtained are reported and discussed.
KW - Emulation of human learning mechanisms
KW - heuristic theory of learning
KW - learning from examples in noisy environments
KW - learning of production rules
KW - machine learning
KW - pattern recognition
UR - http://www.scopus.com/inward/record.url?scp=0024048849&partnerID=8YFLogxK
U2 - 10.1109/34.3917
DO - 10.1109/34.3917
M3 - Article
SN - 0162-8828
VL - 10
SP - 555
EP - 578
JO - IEEE Transactions on Pattern Analysis and Machine Intelligence
JF - IEEE Transactions on Pattern Analysis and Machine Intelligence
IS - 4
ER -