A. Srinivasan and S. Muggleton. Comparing the use of background knowledge by two ilp systems. In L. de Raedt, editor, ILP-95., Katholieke Universiteit Leuven, 1995.  Home/Search   Document Not in Database   Summary   Related Articles   Check

....learning tasks. This process, called propositionalization, has already been investigated within the multipleinstance framework in [8] In this section NAIVERIPPERMI and RIPPERMI will be used in association with REPART [8] to solve a traditional ILP problem : the mutagenesis prediction task [7]. 5.1 Solving the Mutagenesis Problem with a Multiple Instance Learner The mutagenesis prediction problem [7] consists in inducing a theory which can be used to predict whether a molecule is mutagenic or not. To achieve this, a dataset describing 188 molecules with prolog facts is used. Several ....

....framework in [8] In this section NAIVERIPPERMI and RIPPERMI will be used in association with REPART [8] to solve a traditional ILP problem : the mutagenesis prediction task [7] 5. 1 Solving the Mutagenesis Problem with a Multiple Instance Learner The mutagenesis prediction problem [7] consists in inducing a theory which can be used to predict whether a molecule is mutagenic or not. To achieve this, a dataset describing 188 molecules with prolog facts is used. Several relational descriptions of the domain are available. We will use the description termed B 2 [7] where atoms and ....

[Article contains additional citation context not shown here]

A. Srinivasan and S. Muggleton. Comparing the use of background knowledge by two ilp systems. In L. de Raedt, editor, ILP-95., Katholieke Universiteit Leuven, 1995.

....learning tasks. This process, called propositionalization, has already been investigated within the multiple instance framework in [17] In this section NAIVE RIPPERMI and RIPPERMI will be used in association with REPART [17] to solve a traditional ILP problem : the mutagenesis prediction task [15]. 5.1 Solving the Mutagenesis Problem with a Multiple Instance Learner The mutagenesis prediction problem [15] consists in inducing a theory which can be used to predict whether a molecule is mutagenic or not. To achieve this, a dataset describing 188 molecules with prolog facts is used. Several ....

....framework in [17] In this section NAIVE RIPPERMI and RIPPERMI will be used in association with REPART [17] to solve a traditional ILP problem : the mutagenesis prediction task [15] 5. 1 Solving the Mutagenesis Problem with a Multiple Instance Learner The mutagenesis prediction problem [15] consists in inducing a theory which can be used to predict whether a molecule is mutagenic or not. To achieve this, a dataset describing 188 molecules with prolog facts is used. Several relational descriptions of the domain are available. We will use the description termed # # [15] where atoms ....

[Article contains additional citation context not shown here]

A. Srinivasan and S. Muggleton. Comparing the use of background knowledge by two ilp systems. In L. de Raedt, editor, ILP-95., Katholieke Universiteit Leuven, 1995.

.... both the examples and the theories to be learned [14] The application of ILP to problems involving numbers has shown the need for classical knowledge about numerical domains, e.g. the relation less than) This has often been met by supplying the learner with some ad hoc additional knowledge [22]; however, one cannot get rid of the inherent limitations of LP regarding numerical variables: functions are not interpreted, i.e. they act as functors in terms. The consequences for that are detailed in section 2.1. Another possibility consists in mapping an ILP problem into an attribute value ....

....e.g. which suffer from the same limitations. Thus, in order to handle numerical variables without extending unification, one must carefully design predicate definitions, and use the interpretation of functions when some ground terms are found. A clever example of such a definition, found in [22], is reported here. The goal is to define the predicate. First thing is to handle the ground case: X Y : number(X) number(Y) X = Y. X X : number(X) Then, in order to handle the non ground variables, one must introduce explicitly a way to bind the variables. The approach presented in ....

[Article contains additional citation context not shown here]

A. Srinivasan and S. Muggleton. Comparing the use of background knowledge by two ILP systems. In L. de Raedt, editor, Proceedings of ILP-95. Katholieke Universiteit Leuven, 1995.

....by , for j = 1 : j. Finally, the distance between any two examples has complexity O(d Theta K Theta j Theta V 2 ) 4.3 Experimentation This approach is evaluated on the well studied mutagenesis problem [13, 21] Table 4. a) reports the best results obtained by FOIL, PROGOL and STILL [20, 18]. FOIL and PROGOL have been evaluated via 10 fold crossvalidation; STILL was evaluated in a similar way, only including 25 runs (with different random seeds) instead of 10, as recommended for evaluating stochastic processes. Run times (in seconds) are measured on HP 735 workstations. DISTILL is ....

A. Srinivasan and S. Muggleton. Comparing the use of background knowledge by two ILP systems. In L. de Raedt, editor, Proceedings of ILP-95. Katholieke

....in STILL. The algorithms of induction and classification in STILL are given, together with the corresponding polynomial complexity results. Section 4 is devoted to experimental validation on the mutagenicity problem; STILL results compare favorably to those of PROGOL and FOIL, reported from [33]. Finally, some avenues for further research are discussed in section 5. 2 Disjunctive Version Spaces in LP and CLP This section illustrates the Disjunctive Version Space approach on a problem pertaining to organic chemistry [11] the mutagenicity problem is one most famous testbed in ILP [33, ....

....[33] Finally, some avenues for further research are discussed in section 5. 2 Disjunctive Version Spaces in LP and CLP This section illustrates the Disjunctive Version Space approach on a problem pertaining to organic chemistry [11] the mutagenicity problem is one most famous testbed in ILP [33, 10]. A more detailed presentation of Disjunctive Version Spaces in the frame of attribute value and CLP languages can be found in [28] and [30] 2.1 Data and language of examples The mutagenicity problem consists in discriminating organic molecules (nitroaromatic compounds) depending on their ....

[Article contains additional citation context not shown here]

A. Srinivasan and S. Muggleton. Comparing the use of background knowledge by two ILP systems. In L. de Raedt, editor, Proceedings of ILP-95. Katholieke Universiteit Leuven, 1995.

....handling facilities, FORS (Karalic, 1995) An attribute value description of the molecules is also available; this second dataset has been processed by linear regression (LR) neural nets (NN) and CART. Table 3 displays the reference results, obtained by 10 fold crossvalidation and reported from (Srinivasan Muggleton, 1995) and (Karalic, 1995) Table 3: Reference Results LR NN CART PROGOL FOIL FORS Acc. 89 89 88 88 86 89 Sigma 2 2 2 3 3 6 The computational costs given for PROGOL, FOIL and FORS vary with the description used: PROGOL takes from 117; 000 to 40; 000 seconds (on HP735) FOIL from 9; 000 to :5 ....

....and on the mutagenesis problem, STILL shows quite competitive in terms of predictive accuracy. Further, it is faster by two or three orders of magnitude: e.g. PROGOL and FOIL process the purely structural description of molecules (atoms and bonds only) in respectively 60,000 and 9,000 seconds (Srinivasan Muggleton, 1995), whereas STILL takes less than two minutes for the same dataset (these times on HP 735 workstation) 5 Discussion and Perspectives STILL inherits most characteristics of Version Spaces and DiVS, notably the absence of restrictions on candidate hypotheses, except consistency and (partial) ....

Srinivasan, A., and Muggleton, S. 1995. Comparing the use of background knowledge by two ILP systems. In de Raedt, L., ed., Proceedings of ILP-95. Katholieke Universiteit Leuven.

....de ces mol ecules est egalement disponible; ces donn ees ont et e trait ees par les techniques de r egression lin eaire (LR) les r eseaux neuronaux (NN) et le syst eme CART. La figure 3 r ecapitule les meilleurs r esultats obtenus par ces syst emes, evalu es par 10 validation crois ee) [20] et [6] Figure 3: R esultats de r ef erence LR NN CART PROGOL FOIL FORS bien class es 89 89 88 88 86 89 D eviation standard 2 2 2 3 3 6 4.2 Conditions d exp erimentation STILL a et e evalu e par un processus analogue a la 10 validation crois ee, mais faisant intervenir plus d ex ecutions ....

A. Srinivasan and S. Muggleton. Comparing the use of background knowledge by two ILP systems. In L. de Raedt, editor, Proceedings of ILP-95. Katholieke Universiteit Leuven, 1995.

....classifier. Experiments consider the mutagenesis problem [8] which has been thoroughly investigated [19] 4.1 The problem domain The dataset is composed of 125 active molecules and 63 inactive molecules. The background knowledge considered here includes all available information (termed B 4 in [18]) the description of atoms and bonds in the molecules, numerical attributes, and simple chemical concepts (e.g. benzenic or methyl group) involved in the molecules. The reference results obtained by PROGOL, FOIL and STILL on this problem (reported from [18, 17] are given in Table 2. The results ....

....available information (termed B 4 in [18] the description of atoms and bonds in the molecules, numerical attributes, and simple chemical concepts (e.g. benzenic or methyl group) involved in the molecules. The reference results obtained by PROGOL, FOIL and STILL on this problem (reported from [18, 17]) are given in Table 2. The results for STILL are averaged over eight parameter settings (the noise parameter is 0 or 2, and the sparseness parameter ranges in [6,9] Run times (in seconds) are measured on HP 735 workstations. System Accuracy Time FOIL 82 Sigma 3 .5 PROGOL 88 Sigma 2 40 ....

A. Srinivasan and S. Muggleton. Comparing the use of background knowledge by two ILP systems. In L. de Raedt, editor, Proceedings of ILP-95. Katholieke Universiteit Leuven, 1995.

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