Viewpoints are coherent collections of facts that describe a concept from a particular perspective. They are essential for a wide variety of tasks, such as explanation generation and qualitative modeling. We have identified many types of viewpoints and developed a program, the View Retriever, for extracting them from knowledge bases, either singly or in combinations. The View Retriever provides a general solution to the central problem in extracting viewpoints: determining which facts are relevant to requested viewpoints. Our evaluation of the View Retriever indicates that it provides most of the types of viewpoints that people use, and its viewpoints are comparable in coherence to those constructed by people.

Ideally, definitions induced from examples should consist of all, and only, disjuncts that are meaningful (e.g., as measured by a statistical significance test) and have a low error rate. Existing inductive systems create definitions that are ideal with regard to large disjuncts, but far from ideal with regard to small disjuncts, where a small (large) disjunct is one that correctly classifies few (many) training examples. The problem with small disjuncts is that many of them have high rates of misclassification, and it is difficult to eliminate the errorprone small disjuncts from a definition without adversely affecting other disjuncts in the definition. Various approaches to this problem are evaluated, including the novel approach of using a bias different than the "maximum generality" bias. This approach, and some others, prove partly successful, but the problem of small disjuncts remains open.

Ideally, definitions induced from examples should consist of all, and only, disjuncts that are meaningful (e.g., as measured by a sta­ tistical significance test) and have a low error rate. Existing inductive systems create defini­ tions that are ideal with regard to large dis­ juncts, but far from ideal with regard to small disjuncts, where a small (large) disjunct is one that correctly classifies few (many) training ex­ amples. The problem with small disjuncts is that many of them have high rates of misclassi- fication, and it is difficult to eliminate the error- prone small disjuncts from a definition without adversely affecting other disjuncts in the defi­ nition. Various approaches to this problem are evaluated, including the novel approach of us­ ing a bias different than the "maximum gen­ erality" bias. This approach, and some oth­ ers, prove partly successful, but the problem of small disjuncts remains open.

supervised discrimination of clustered data via optimization of binary information gain. Induction over the unexplained: Integrated learning of concepts with both explainable and conventional aspects.mann machines: Constraint satisfaction networks that learn.gation learning for multi-layer feed-forward neural networks using the conjugate gradient method. ARTMAP: Supervised real-time learning and classiication of nonstationary data by a self-organizing neural network. Andrea Pohoreckyj Danyluk. Finding new rules for incomplete theories: explicit biases for induction with contextual information. International application of a new probability algorithm for the diagnosis of coronary artery disease. Transfer in neural networks 35 critical to learning systems of the future, which will require the ability to quickly adapt to new situations based on past experience. for past input into this research program. Tom Fawcett helped with many of the symbolic references. Alex Waibel suggested the application of transfer to speaker-dependent tasks. The heart disease data was collected by: beach and Cleveland Clinic Foundation. We used software distributed with McClelland and Rumelhart, 1988 ] for many of our simulations. A neural-net training program based on conjugate-gradient optimization. 0 0 0 0 1 1 0 1 1 1 1 0 0 0 0 1 1 0 0 0 0 1 1 0 1 1 0 1 1 1 0 0 1 0 0 1 Source 1 Feature 1 HU 1 HU1 HU 2 0 0 0 0 1 1 0 1 1 1 1 0 0 0 0 1 1 0 0 0 0 1 1 0 1 1 0 1 1 1 0 0 1 0 0 1 HU 1 HU1 HU 2 HU 3 0 0 0 0 1 1 0 1 1 1 1 0 0 0 0 1 1 0 0 0 0 1 1 0 1 1 0 1 1 1 0 0 1 0 0 1 Source 3