Ortega, J. (1995). On the informativeness of the DNA promoter sequences domain theory. Journal of Artificial Intelligence Research, 2, 361--367.  Home/Search   Document Details and Download   Summary   Related Articles   Check

.... (Mahoney Mooney, 1994; Mahoney, 1996; Buntine, 1991; Lam Bacchus, 1994; Russell, Binder, Koller, Kanazawa, 1995; Ramachandran Mooney, 1998) or, most relevant to this paper, by interpreting a logical theory in a probabilistic manner (Towell Shavlik, 1993; Koppel, Feldman, Segre, 1994b; Ortega, 1995). All of these methods depend on the use of a sufficiently large set of training examples for diagnosing and repairing flaws in the theory. In this paper, we offer a universal method of theory reinterpretation that makes only marginal use of training examples. In the simplest version of the ....

....as proved and those scoring below as unproved, we get a classification accuracy of 93.4 . This example illustrates how by softening a theory we may dramatically improve its classification accuracy (50 to 93.4 ) without doing any revision whatsoever. This merely confirms earlier results, such as (Ortega, 1995) that indicate that simple numerical generalization strategies are very effective for this particular theory. It is surprising, though, that so simple a scheme achieves as good or better results on this theory than many theory revision techniques (Towell Shavlik, 1993; Koppel et al. 1994a; ....

[Article contains additional citation context not shown here]

Ortega, J. (1995). On the informativeness of the DNA promoter sequences domain theory. Journal of Artificial Intelligence Research, 2, 361--367.

....function. If no value for m is explicitly stated, we assume m = n. It should be noted that despite the possible range for values of m, only m close to n make sense. If m n (i.e. m is substantially smaller than n) then a loose matching criteria are defined. The experiments presented in [Ort95] show that setting m = n Gamma 2 allowed for the best results considering the ratio of correct and incorrect answers. Using the previous servo case example, cardinality relaxation means that we do not require all four attributes to match (motor, screw, pgain and vgain) but only, for example 2 of ....

.... simple cross validation (leave one out method) and we computed significance intervals and variances from the averaging 20 random trials (see Table 4) In TA3 Robot 1 only cardinality relaxation was used (i.e. only m of n, m n, attributes were required to match) A similar approach presented in [Ort95] yielded analogous results accuracy improvement for m n. TA3 Robot 2 used value relaxation (i.e. relaxing attribute values) first and if this failed then cardinality relaxation was performed. It is interesting to note that TA3 Robot 1 outperformed TA3 Robot 2 , although it uses a domain ....

J. Ortega. On the informativeness of the DNA promoter sequences domain theory. Journal of Artificial Intelligence Research, 2:361--367, 1995. Research Note.

....collecting cases sharing some features, for example using an IVF case [JS95] all women that have pregnancy with complications are in their third or fourth cycle. However, the pattern identification is not sufficient patterns need to be described: given a set of cases labeled by class. In [Ort95] it was shown that m of n matches allow for improved performance if m is reasonably selected. Our approach of representing cases as sets of categories, each comprising a set of tuples, allows for multiple levels of m of n matching. Thus, important attributes may require n of n matches for a given ....

....consists of 20,000 instances, described by 17 attributes. Based on simple crossvalidation, the basic TA3 system achieved comparable classification accuracy to other systems. Attributes were grouped into categories for selective cardinality relaxation restriction, i.e. selective m of n matching [Ort95]. We have tested several relaxation strategies and attribute grouping approaches that yielded various accuracy results, using 95 confidence intervals. We have also tested the effect of the case base size on the classification accuracy achieved for individual situations. Method Accuracy (20K ....

Ortega, J. (1995). On the informativeness of the DNA promoter sequences domain theory. Journal of Artificial Intelligence Research , 2:361-367. Research Note.

....Performance degradation of IB1 with the number of irrelevant attributes can be solved either by data pre processing (removing all irrelevant attributes) or by the use of an intelligent, selective partial matching algorithm. The latter approach is similar to m of n concepts in machine learning [30], i.e. the system considers only m n attributes during retrieval. There are various methods used to decide m and Ortega [30] presents an evaluation of system performance for various m. In addition to deciding the size of m, the approach described in this paper allows for specifying which ....

....all irrelevant attributes) or by the use of an intelligent, selective partial matching algorithm. The latter approach is similar to m of n concepts in machine learning [30] i.e. the system considers only m n attributes during retrieval. There are various methods used to decide m and Ortega [30] presents an evaluation of system performance for various m. In addition to deciding the size of m, the approach described in this paper allows for specifying which attributes to exclude and for posing additional constraints on attribute values, if desired. CB1 [15] is a PAC (Probably ....

[Article contains additional citation context not shown here]

J. Ortega. On the informativeness of the DNA promoter sequences domain theory. Journal of Artificial Intelligence Research, 2:361--367, 1995.

....number of irrelevant attributes. Such a performance degradation of IB1 can be solved either by data preprocessing that wouldremove all irrelevant attributes or by the use of an intelligent, selective partial matching algorithm. The latter approach is similar to m of n concepts in machine learning [35], i.e. the system considers only m n attributes during retrieval. There are various methods used to decide m and Ortega [35] presents an evaluation of system performance for various m. In addition to deciding the size of m, the approach described in this paper allows for specifying which ....

....all irrelevant attributes or by the use of an intelligent, selective partial matching algorithm. The latter approach is similar to m of n concepts in machine learning [35] i.e. the system considers only m n attributes during retrieval. There are various methods used to decide m and Ortega [35] presents an evaluation of system performance for various m. In addition to deciding the size of m, the approach described in this paper allows for specifying which attributes to exclude and for posing additional constraints on attribute values, if desired. CB1 [18] is a PAC (Probably ....

J. Ortega. On the informativeness of the DNA promoter sequences domain theory. Journal of Artificial Intelligence Research, 2:361--367, 1995.

.... (Mylopoulos et al. 1990) representation language allows for treating objects (cases) and their attributes uniformly and thus diminishing problems discussed earlier (Dierbach Chester 1991) In addition, it allows for lowering the impact of irrelevant attributes on system performance (Aha 1992; Ortega 1995). Categories allow for improved system performance (Jurisica Glasgow 1995; Jurisica Shapiro 1995; Jurisica Glasgow 1996a) Knowledge discovery techniques are used to organize case attributes into categories (Jurisica 1996; Jurisica et al. 1996) The main goal is to retrieve only useful ....

....range defined in the context. For a domain value constraint, the attribute value in the source case must be an instance of the attribute value in the context. 2. Cardinality constraints specify the number of attributes required to match for a particular category, similarly as m of n matching in (Ortega 1995). Exact specifies that all attributes in the category of a source case and the interpretation of the target case must match. Some(m) determines that not all attributes in a category are required to match, but at least m of n attributes must match the context. It should be noted that Exact = ....

Ortega, J. 1995. On the informativeness of the DNA promoter sequences domain theory. Journal of Artificial Intelligence Research 2:361--367. Research Note.

....this idea has great intuitive appeal, revision is not always the best way to use a given theory. Another class of methods does not attempt to repair the given theory, but to reinterpret it in a more profitable manner. This can be done by treating it in a probabilistic manner (Koppel et al. 1994b; Ortega, 1995), or by using the theory as a resource for constructive induction (Donoho and Rendell, 1995; Ortega and Fisher, 1995) Each class of methods, either revision or reinterpretation, works well for some problems and less well for others. This is entirely expected, since any particular type of learning ....

....to use this information well. A better idea to do in this case is to reinterpret the theory more flexibly. For example, results on the promoter theory show better classification accuracy using probabilistic reinterpretation than by using revision (Koppel et al. 1994b; Donoho and Rendell, 1995; Ortega, 1995). Typically, we do not know at the outset what the nature of the problems with a given theory are. This paper describes a meta level learning algorithm, which determines what kind of learning algorithm is most appropriate given a particular flawed theory. The algorithm uses a set of ....

[Article contains additional citation context not shown here]

J. Ortega. 1995. On the informativeness of the dna promoter sequences domain theory. Journal of Artificial Intelligence Research, 2:361--367.

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