Lavrac, N. and Dzeroski, S. "Background knowledge and declarative bias in inductive concept learning" In Jantke, K., editor, Proc. Third International Workshop on Analogical and Inductive Inference, pp 51-71. Springer, Berlin.  Home/Search   Document Not in Database   Summary   Related Articles   Check

....of literals in the body is bounded by a constant j, and the depth of the clause is bounded by a constant i. The ij determinacy restriction was first used in the GOLEM system [Muggleton and Feng, 1992] and has since been incorporated in several other practical learning systems [Cohen, 1993a; Lavrac and Dzeroski, 1992; Quinlan, 1991] Some very recent work [Kietz, 1993] shows that a single clause is not pac learnable if the ij determinacy condition does not hold; specifically, it is shown that neither the language of indeterminate clauses of fixed depth nor the language of determinate clauses of arbitrary ....

.... Horn clause, or (against simple distributions) a nonrecursive program containing k such clauses [D zeroski et al. 1992] The latter result is of interest because the property of ij determinacy is made use of by several practical learning systems [Muggleton and Feng, 1992; Cohen, 1993a; Lavrac and Dzeroski, 1992; Quinlan, 1991] and is also used extensively in this paper: indeed, this paper can be considered an investigation of the degree to which this language can be generalized while preserving pac learnability. Some previous negative results also exist. There are a number of results on the ....

Nada Lavrac and Saso Dzeroski. Background knowledge and declarative bias in inductive concept learning. In K. P. Jantke, editor, Analogical and Inductive Inference: International Workshop AII'92. Springer Verlag, Daghstuhl Castle, Germany, 1992. Lecture in Artificial Intelligence Series #642.

....of depth two. zero(B) one(C) decrement(B,D) decrement(A,E) multiply(B,C,G) divide(G,A,F) choose(E,D,F) The program GOLEM [Muggleton and Feng, 1992] learns ij determinate programs, and related restrictions have been adopted by several other practical learning systems [Quinlan, 1991; Lavrac and Dzeroski, 1992; Cohen, 1993d] The learnability of ij determinate clauses has also received some formal study, which we will review in Section 6. 2.2.3 Mode constraints and declarations We define the mode of a literal L appearing in a clause C to be a string s such that the initial character of s is the ....

Nada Lavrac and Saso Dzeroski. Background knowledge and declarative bias in inductive concept learning. In K. P. Jantke, editor, Analogical and Inductive Inference: International Workshop AII'92. Springer Verlag, Daghstuhl Castle, Germany, 1992. Lectures in Artificial Intelligence Series #642.

....knowledge one can consider nondeleting graph productions, that have to be applied to the training examples and new objects to be classified in a first step. An example for the context dependent classification of edges is the definition of the concept grandmother in terms of father and mother ([7]) The LINUS approach and our approach both try to describe the problem by feature vectors and use decision tree induction. The classification of structured objects may be applicable to the problem of mutagenicity of chemicals described in [6] 8 Concluding Remarks The paper presented the ....

N. Lavrac and S Dzeroski. Background Knowledge and Declarative Bias in Inductive Concept Learning. In K.P. Jantke, editor, Analogical and Inductive Inference, Proc. Int. Workshop AII'92, number 642 in Lecture Notes in Artificial Intelligence, pages 51--71. Springer Verlag, 1992.

....to ILP is to first generate logic program clauses, and then use these clauses convert the examples and background knowledge into a propositional form. One can then apply conventional propositional learning methods to the constructed dataset. A number of theoretically proposed learning methods [ Lavrac and Dzeroski, 1992; Dzeroski et al. 1992; Cohen, 1994b ] and at least one practical learning system (LINUS [ Lavrac and Dzeroski, 1994 ] rely on this technique. We evaluated the following simple indirect learning system. Given a dataset, a set of background relations, and a set of suggested modes, we enumerated ....

Nada Lavrac and Saso Dzeroski. Background knowledge and declarative bias in inductive concept learning. In K. P. Jantke, editor, Analogical and Inductive Inference: International Workshop AII'92. Springer Verlag, Daghstuhl Castle, Germany, 1992. Lectures in Artificial Intelligence Series #642. 26

....models in (Ali Pazzani, 1995) 130 K. ALI AND M. PAZZANI 3. Ali Pazzani (1995) presents details on how to deal with recursive concepts in the single, most reliable rule framework. 4. King Rook King is a fully determinate domain so it can be converted into attribute value form as is done by Lavrac Dzeroski (1992). However, in this paper, that knowledge is not utilized by the learning programs so the domain has to be treated as a relational domain. FOIL and HYDRA can also run on non determinate domains. 5. x class noise means that the class assignments of x of the examples were randomly reassigned for ....

Lavrac, N. & Dzeroski, S. (1992.) Background knowledge and declarative bias in inductive concept learning. In Oantke, K., Proceedings of the Third International Workshop on Analogical and Inductive Inference. Berlin, Germany: Springer.

....predicate being learned (to allow for recursive definitions) In general, the size of this hypotheses space turns out to be huge. In order to discover a correct hypothesis feasibly, this space must be structured and searched efficiently. Many ILP systems provide extra information or meta knowledge [8, 10] about the relation being learned. This information renders a large number of hypotheses in the search space incompatible, and the system can safely exclude them from the search space. This compacts or biases the search space and improves the efficiency of the ILP system. Current ILP systems ....

Nada Lavrac and Saso Dzeroski. Background knowledge and declarative bias in inductive concept learning. In K. P. Jantke, editor, Analogical and Inductive Inference, Lecture Notes in Artificial Intelligence. Springer-Verlag, October 1992.

....huge. In order to discover a correct hypothesis feasibly, this space must be structured and searched Supported by an Australian Postgraduate Research Award. y Supported by a grant from the Australian Research Council. efficiently. Many ILP systems provide extra information or meta knowledge [6, 8] about the relation being learned. This information renders a large number of hypotheses in the search space incompatible, and the system can safely exclude them from the search space. This compacts or biases the search space and improves the efficiency of the ILP system. Table 1 shows the biasing ....

Nada Lavrac and Saso Dzeroski. Background knowledge and declarative bias in inductive concept learning. In K. P. Jantke, editor, Analogical and Inductive Inference, Lecture Notes in Artificial Intelligence. Springer-Verlag, October 1992.

....from the head of a clause to appear in the literals in clause body (constrained clauses) using determinacy allows for a restricted form of new local variables to be introduced in the body of an induced clause. This approach to learning of determinate clauses is implemented in the system DINUS [9, 25, 26]. Compared to the LINUS algorithm outlined in Figure 1, the DINUS learning algorithm consists of the same main steps. Steps 2 and 4, outlined below, are more elaborate in DINUS. ffl In step 2, the ILP problem is transformed to propositional form as follows. The algorithm first constructs, ....

....T B . ffl In step 4, the induced propositional description is transformed back to a set of determinate DHDB clauses, by adding the necessary determinate literals which introduced the new variables. A more detailed description of the algorithm and examples of its performance can be found in [25, 26]. 19 4 Extending LINUS to learn non determinate clauses In this section we describe how the DINUS restriction of determinate literals can be overcome, by employing a so called individual centred representation. We start by discussing feature construction as an important, possible separate step, ....

N. Lavrac and S. Dzeroski. Background knowledge and declarative bias in inductive concept learning. In K. Jantke, editor, Proceedings 3rd International Workshop on Analogical and Inductive Inference, pages 51--71. SpringerVerlag, 1992. (Invited paper).

No context found.

Lavrac, N. and Dzeroski, S. "Background knowledge and declarative bias in inductive concept learning" In Jantke, K., editor, Proc. Third International Workshop on Analogical and Inductive Inference, pp 51-71. Springer, Berlin.

No context found.

Nada Lavrac and Saso Dzeroski. Background knowledge and declarative bias in inductive concept learning. In K. P. Jantke, editor, Analogical and Inductive Inference: International Workshop AII'92. Springer Verlag, Daghstuhl Castle, Germany, 1992. Lectures in Artificial Intelligence Series #642.

No context found.

Nada Lavrac and Saso Dzeroski. Background knowledge and declarative bias in inductive concept learning. In K. P. Jantke, editor, Analogical and Inductive Inference: International Workshop AII'92. Springer Verlag, Daghstuhl Castle, Germany, 1992. Lectures in Artificial Intelligence Series #642.

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