Extracting rules from RBFs is not a trivial task because of nonlinear functions or high input dimensionality. In such cases, some of the hidden units of the RBF network have a tendency to be "shared" across several output classes or even may not contribute to any output class. To address this we have developed an algorithm called LREX (for Local Rule EXtraction) which tackles these issues by extracting rules at two levels: hREX extracts rules by examining the hidden unit to class assignments while mREX extracts rules based on the input space to output space mappings. The rules extracted by our algorithm are compared and contrasted against a competing local rule extraction system. The central claim of this paper is that local function networks such as radial basis function (RBF) networks have a suitable architecture based on Gaussian functions that is amenable to rule extraction.

Feature selection is a common and key problem in many classification and regression tasks. It can be viewed as a multiobjective optimisation problem, since, in the simplest case, it involves feature subset size minimisation and performance maximisation. This paper presents a multiobjective evolutionary approach for feature selection. A novel commonality-based crossover operator is introduced and placed in the multiobjective evolutionary setting. This specialised operator helps to preserve building blocks with promising performance. Selection bias reduction is achieved by resampling. We argue that this is a generic approach, which can be used in many modelling problems. It is applied to feature selection on different neural network architectures. Results from experiments with benchmarking data sets are given.

This paper discusses the use of multicriteria genetic algorithms for feature selection in classification problems. This feature selection approach is shown to yield a diverse population of alternative feature subsets with various accuracy/complexity trade-off. The algorithm is applied to select features for performing classification with fuzzy models, and is evaluated on two real-world data sets. We discuss when multicriteria genetic algorithm feature selection is preferable to a sequential feature selection procedure, namely backwards elimination. Among the key features of the presented approach are its computational simplicity, effectiveness on real world problems and the potential it has to become a powerful tool aiding many empirical modeling and data mining processes.

This paper deals with an evaluation and comparison of the accuracy and complexity of symbolic rules extracted from radial basis function networks and multilayer perceptrons. Here we examine the ability of rule extraction algorithms to extract meaningful rules that describe the overall performance of a particular network. In addition, the paper also highlights the suitability of a specific neural network architecture for particular classification problems. The study carried out on the extracted rule quality and complexity also has a direct bearing on the use of rule extraction algorithms for data mining and knowledge discovery.

In [3] Friedberg showed that every Turing degree ≥ 0′ is the jump of some degree. Using the relativized version of this theorem it can be shown by finite induction that if a ≥ 0(n)0(n) then there is a b such that b(n) = a (our notation is defined in §1). It is natural to ask whether these results can be extended into the transfinite. Is it true for example, that whenever a ≥ 0(ω) there is a b such that = b(ω)= a? In §2 we use forcing to prove this result. (The usefulness of forcing in answering questions involving the jump operation on degrees of unsolvability has been previously demonstrated in Selman [5] where, for example, forcing was used to construct degrees a and b such that a(ω) = b(ω) = a ∪ b = 0(ω).) In §3 we generalize the methods of §2 to show that if α is a recursive ordinal and a ≥ 0(α) then there is a bsuch that b(α) = a, i.e. the Friedberg result can be extended to all recursive ordinal levels. Thomason [6] used a forcing argument to show: If (the Kleene set of notations for the recursive ordinals) then there is a B such that (the set of notations for ordinals recursive in B). In §4 we show this result holds when hyperarithmetic reducibility is replaced by Turing reducibility: If then there is a B such that .

Let α be an admissible ordinal and let L be the first order language with equality and a single binary relation ≤. The elementary theory of the α-degrees is the set of all sentences of L which are true in the universe of the α-degrees when ≤ is interpreted as the partial ordering of the α-degrees. Lachlan [6] showed that the elementary theory of the ω-degrees is nonaxiomatizable by proving that any countable distributive lattice with greatest and least members can be imbedded as an initial segment of the degrees of unsolvability. This paper deals with the extension of these results to α-recursion theory for an arbitrary countable admissible α > ω. Given α, we construct a set A with α-degree a such that every countable distributive lattice with greatest and least member is order isomorphic to a segment of α-degrees {d ∣ a ≤αd≤αb} for some α-degree b. As in [6] this implies that the elementary theory of the α-degrees is nonaxiomatizable and hence undecidable. A is constructed in §2. A is a set of integers which is generic with respect to a suitable notion of forcing. Additional applications of such sets are summarized at the end of the section. In §3 we define the notion of a tree and construct a particular tree T0 which is weakly α-recursive in A. Using T0 we can apply the techniques of [6] and [2] to α-recursion theory. In §4 we reduce our main results to three technical lemmas concerning systems of trees. These lemmas are proved in §5.

This paper investigates the problem of extending the recursion theoretic construction of a minimal degree to the Kripke [2]-Platek [5] recursion theory on the ordinals less than an admissible ordinal α, a theory derived from the Takeuti [11] notion of a recursive function on the ordinal numbers. As noted in Sacks [7] when one generalizes the recursion theoretic definition of relative recursiveness to α-recursion theory for α > ω the two usual definitions give rise to two different notions of reducibility. We will show that whenever α is either a countable admissible or a regular cardinal of the constructible universe there is a subset of α whose degree is minimal for both notions of reducibility. The result is an excellent example of a theorem of ordinary recursion theory obtainable via two different constructions, one of which generalizes, the other of which does not. The construction which cannot be lifted to α-recursion theory is that of Spector [10]. We sketch the reasons for this in §3.