T.G. Dietterich & R.S. Michalski, Inductive learning of structural description: evaluation criteria and comparative review of selected methods, Artificial Intelligence, 16(3), 1981, 257--294.  Home/Search   Document Not in Database   Summary   Related Articles   Check

....using a fixed amount of memory and application specific rules for pruning nonpromising search alternatives. It was experimentally observed that beam search can significantly reduce the computation of its underlying search method [4] Beam search has been successfully applied to many problems [11, 12, 29, 35, 40]. Since both state space reduction and beam search use heuristic node pruning, the former can be regarded as an extension of the latter. The main contribution of this research to beam search is twofold. First, we extended the idea of heuristic node pruning to a depth first search algorithm and ....

G. Dietterich and R. S. Michalski. Inductive learning of structural descriptions: Evaluation criteria and comparative review of selected methods. Artificial Intelligence, 16:257--294, 1981.

....have implemented a system which automatically generates examples for a constraint stated in the theory and we have experimented with the system in the BoyerMoore theorem prover. 1.1 An Overview Examples are, in general, a very useful tool in Artificial Intelligence. Many machine learning systems [52, 33, 10, 14, 32] use examples for the tasks of generating concepts and conjectures. For instance, Winston s system [52] learns structural descriptions from examples. His system is presented with training instances positive examples and near misses of a structure (concept) to be learned such as an arch and ....

Dietterich, T. G. and Michalski, R. S. "Inductive Learning of Structural Descriptions: Evaluation Criteria and Comparative Review of Selected Methods". Artificial Intelligence 16 (1981), 257-294.

.... researchers (Matheus, 1991; Aha, 1991; Kadie, 1991; Ragavan et al. 1993) Constructing new feature by hand is often difficult (Quinlan, 1983) The goal of constructive induction is to automatically transform the original representation space into a new one where the regularity is more apparent (Dietterich Michalski, 1981; Mehra et al. 1989) thus yielding improved classification accuracy. Several machine learning algorithms perform feature construction by extending greedy, hill climbing strategies, including FRINGE (Pagallo, 1989) GREEDY3 (Pagallo Haussler, 1990) DCFringe (Yang et al. 1991) CITRE ....

Dietterich, T. G. & Michalski, R. S. "Inductive Learning of Structural Description : Evaluation Criteria and Comparative Review of Selected Methods", Artificial Intelligence 16 (3), p257-294, 1981.

.... need for useful new features has been suggested by many researchers (Matheus, 1991; Ragavan et al. 1993) Constructing new features by hand is often difficult (Quinlan 1993) The goal of constructive induction is to automatically generate 1 new features so that the regularity is more apparent (Dietterich Michalski, 1981), thus yielding improved classification accuracy. The constructed features should make concept learning easier for many learning algorithms. There are currently many constructive induction algorithms based on the strategy of constructing new attributes, including FRINGE (Pagallo, 1989) GREEDY3 ....

Dietterich, T. G. & Michalski, R. S. "Inductive Learning of Structural Description : Evaluation Criteria and Comparative Review of Selected Methods ", Artificial Intelligence 16 (3), p257-294, 1981.

....only if it is not satisfied by any of the negative examples. The learned concept is an admissible rule if and only if it is both characteristic and discriminant [DiM83,GeN87] Most learning algorithms are designed for learning admissible rules [DiM83,Mic83] A few algorithms, such as INDUCE 1. 2 [DiM81] and SPROUTER [HaM77] are designed for learning characteristic rules. DBROUGH [HuC94a, HuC94b, HSCZ94, HCH94, HCS94] can discover characteristic rules, discriminant rules and some other knowledge rules. 2.1.4 Control Strategies in Learning from Examples Induction methods can be divided into ....

....the search. These methods search a set of possible generalisations in an attempt to find a few best hypotheses that satisfy certain requirements. Typical examples of systems which adopt this strategy are AM [Len77] DENDRAL and Meta DENDRAL [BuM78] and the approach used in the INDUCE system [DiM81]. Data driven techniques generally have the advantage of supporting incremental learning. The learning process can start not only from the specific training examples, but also from the rules which have already been discovered. The learning systems are capable of updating the existing hypotheses ....

T.G. Dietterich and R.S. Michalski, (1981). Inductive Learning of Structural Descriptions: Evaluation Criteria and Comparative Review of Selected Methods, Artificial Intelligence, Vol. 16, 251-294.

....it is a possible explanation. The point is that XPs are associated with stereotypical situations and people in memory. An understander needs to learn the stereotypical categories that serve as useful indices for motivational explanations. This is a type of inductive category formation [Diettrich and Michalski, 1981]; however, the generalization process is constrained so that the features selected for generalization are those that are causally relevant to the explanations being indexed [Flann and Dietterich, 1989] AQUA indexes motivational XPs in memory using typical contexts in which the XPs might be ....

T. G. Diettrich and R. S. Michalski. Inductive Learning of Structural Descriptions: Evaluation Criteria and Comparative Review of Selected Methodologies. Artificial Intelligence, 16:257--294, 1981.

....this might indicate random data. ffl Another advantage is the fact that all of the training data can be used for inducing and evaluating theories. Alternative methods like cross validation or the use of some arbitrary quality criteria (e.g. the lexicographical evaluation function LEF of [Dietterich Michalski 81] need to split the training set into two parts: one will be used for induction, the other one will be used for evaluation. Splitting can be problematic for small training sets: only regularities present in both sub sets can be found and kept. Splitting of small training sets may cause an ....

....axis parallel nested hyper rectangles. Two different problems may cause irregular distributions of learning examples in the original representation space: noise and or an inadequate description language. As a remedy for the latter problem constructive induction has been introduced, e.g. in [Dietterich Michalski 81] and [Mehra et al. 89] The basic idea is to somehow transform the original representation space into a space where the learning examples exhibit (more) regularities. Usually this is done by introducing new attributes and forgetting old ones. So constructive induction is searching for an adequate ....

Dietterich T.G., Michalski R.S.: Inductive Learning of Structural Descriptions: Evaluation Criteria and Comparative Review of Selected Methods, Artificial Intelligence, 16(3), pp.257-294, 1981.

....axis parallel nested hyper rectangles. Two different problems may cause irregular distributions of learning examples in the original representation space: noise and or an inadequate description language. As a remedy for the latter problem constructive induction has been introduced, e.g. in [Dietterich Michalski 81] and [Mehra et al. 89] The basic idea is to somehow transform the original representation space into a space where the learning examples exhibit (more) regularities. Usually this is done by introducing new attributes and forgetting old ones. So constructive induction is searching for an adequate ....

Dietterich T.G., Michalski R.S.: Inductive Learning of Structural Descriptions: Evaluation Criteria and Comparative Review of Selected Methods, Artificial Intelligence, 16(3), 257-294, 1981.

....axis parallel nested hyper rectangles. Two different problems may cause irregular distributions of learning examples in the original representation space: noise and or an inadequate description language. As a remedy for the latter problem constructive induction has been introduced, e.g. in [Dietterich Michalski 81] and [Mehra et al. 89] The basic idea is to somehow transform the original representation space into a space where the learning examples exhibit (more) regularities. Usually this is done by introducingnew attributes and forgetting old ones. So constructive induction is searching for an adequate ....

Dietterich T.G., Michalski R.S.: Inductive Learning of Structural Descriptions: Evaluation Criteria and Comparative Review of Selected Methods, Artificial Intelligence, 16(3), pp.257-294, 1981.

....description language. Both phenomena lead to complex, convoluted induced concept descriptions which will be hard to understand and will perform poorly at predicting concept membership of unclassified examples. As a remedy for the latter problem constructive induction has been introduced, e.g. in [Dietterich Michalski 81] and [Mehra et al. 89] The basic idea is to somehow transform the original representation space into a space where the learning examples exhibit (more) regularities. Usually this is done by introducing new attributes and forgetting old ones. So constructive induction is searching for an ....

....constructs the opposite attribute more than a certain number of attribute values (typically 5) are out of the healthy range , which is well suited for characterizing people exhibiting serious health problems. 4. 3 Inductive Logic Programming Exercises Encouraged by the original INDUCE system [Dietterich Michalski 81] which was able to learn structural descriptions from examples, and by the current success of LINUS [Dzeroski Lavrac 91] which essentially translates ILP problems into an attribute value representation for efficient induction, we started to examine two classical ILP exercises: illegal ....

Dietterich T.G., Michalski R.S.: Inductive Learning of Structural Descriptions: Evaluation Criteria and Comparative Review of Selected Methods, Artificial Intelligence, 16(3), 257-294, 1981.

....in which it is likely to be useful. As described earlier, XPs are associated with stereotypical situations and people in memory. An understander needs to learn the stereotypical categories that serve as useful indices for volitional explanations. This is a type of inductive category formation [Dietterich and Michalski, 1981]; however, the generalization process is constrained so that the features selected for generalization are those that are causally relevant to the explanations being indexed [Barletta and Mark, 1988; Flann and Dietterich, 1989] XPs are indexed in memory using stereotypical descriptions of the ....

T. G. Dietterich and R. S. Michalski. Inductive Learning of Structural Descriptions: Evaluation Criteria and Comparative Review of Selected Methodologies. Artificial Intelligence, 16:257--294, 1981.

No context found.

Dietterich, T.G. and Michalski, R.S., Inductive learning of structural descriptions: Evaluation criteria and comparative review of selected methods, Artificial bttelligence 16 (1981) 257-294.

No context found.

T.G. Dietterich & R.S. Michalski, Inductive learning of structural description: evaluation criteria and comparative review of selected methods, Artificial Intelligence, 16(3), 1981, 257--294.

No context found.

T.G. Dietterich and R.S. Michalsky. "Inductive Learning of Structural Descriptions: Evaluation Criteria and Comparative Review of Selected Methods". Artificial Intelligence 16, pp. 257-294, 1981.

Online articles have much greater impact   More about CiteSeer.IST   Add search form to your site   Submit documents   Feedback  

CiteSeer.IST - Copyright Penn State and NEC