N. Japkowicz and H. Hirsh. Towards a bootstrapping approach to constructive induction. In Working Notes of the Workshop on Constructive Induction and Change of Representation, 1994.  Home/Search   Document Details and Download   Summary   Related Articles   Check

....that one or two simple low cost tests (e.g. service and or flag) can reduce the number of high cost feature tests from 9 (see Table 4) to at most 3. Our preliminary experiments have thus far confirmed this result. 5 Related Work Our feature construction approach is similar to the work in [JH94]. Our operators are also based on extracted patterns of existing sets of features, and we as well consider the syntactic form of the patterns. We can use fewer and simpler operators, however, since the patterns carry stronger information (e.g. the invariant behavior of an intrusion) We also ....

N. Japkowicz and H. Hirsh. Towards a bootstrapping approach to constructive induction. In Working Notes of the Workshop on Constructive Induction and Change of Representation, 1994.

....suggest that one or two simple low cost tests (e.g. service and or flag) can reduce the number of high cost feature tests from 9 (see Table 4) to at most 3. Our preliminary experiments have thus far confirmed this result. 5 Related Work Our feature construction approach is similar to the work in [JH94]. Our operators are also based on extracted patterns of existing sets of features, and we as well consider the syntactic form of the patterns. We can use fewer and simpler operators, however, since the patterns carry stronger information (e.g. the invariant behavior of an intrusion) We also ....

N. Japkowicz and H. Hirsh. Towards a bootstrapping approach to constructive induction. In Working Notes of the Workshop on Constructive Induction and Change of Representation, 1994.

....in a non degenerate sense and yet they are typically described as selective and thus non constructive (see 1 In the case of C4.5 the intermediate constructs are the internal decision tree nodes while in the case of backpropagation the constructs are the hidden units in the network. [8,9,10,11,12] for a selection of views) The decision criterion that is being applied in such cases is not completely clear. But it appears to be grounded in an evaluation of the simplicity of the partitioning introduced by the feature. Constructs that introduce simple, local partitions (e.g. the ....

....rules. Matheus [27] describes the CITRE system which, to a first approximation, tries to capitalize on the presence of disjunctive regions in decision tree descriptions. Other systems involve a search (i.e. operator based) process working with some sort of relational description language, e.g. [28, 11,9]. In several recent cases, this type of approach has focussed on what are known as counting or M of N features, i.e. features which effectively count the number of occurrences of a particular variable value, cf. 29,30,2,8 31,32] Equating constructive induction with relational learning is also ....

Japkowicz, N. and Hirsh, H. (1994). Towards a bootstrapping approach to constructive induction. Proceedings of ML-COLT'94.

....not directly involved in production of output) that identify partitions on the representation space. 1 These methods would seem to satisfy the criterion of being feature generators in a non degenerate sense and yet they are typically described as selective and thus non constructive (see [9,10,4,11,12] for a selection of views) The decision criterion that is being applied in such cases is not completely clear. But it appears to be grounded in an evaluation of the simplicity of the partitioning introduced by the feature. Constructs that introduce simple, local partitions (e.g. the ....

....rules. Matheus [28] describes the CITRE system which, to a first approximation, tries to capitalize on the presence of disjunctive regions in decision tree descriptions. Other systems involve a search (i.e. operator based) process working with some sort of relational description language, e.g. [29, 11,10]. In several recent cases, this type of approach has focussed on what are known as counting or M of N features, i.e. features which effectively count the number of occurrences of a particular variable value, cf. 30,31,2,9 32,33] It is worth noting, finally, that although all these approaches ....

Japkowicz, N. and Hirsh, H. (1994). Towards a bootstrapping approach to constructive induction. Proceedings of ML-COLT'94.

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