Neural networks, despite their empirically-proven abilities, have been little used for the refinement of existing knowledge because this task requires a three-step process. First, knowledge in some form must be inserted into a neural network. Second, the network must be refined. Third, knowledge must be extracted from the network. We have previously described a method for the first step of this process. Standard neural learning techniques can accomplish the second step. In this paper, we propose and empirically evaluate a method for the final, and possibly most difficult, step. This method efficiently extracts symbolic rules from trained neural networks. The four major results of empirical tests of this method are that the extracted rules: (1) closely reproduce (and can even exceed) the accuracy of the network from which they are extracted; (2) are superior to the rules produced by methods that directly refine symbolic rules; (3) are superior to those produced by previous techniques fo...

Hybrid learning methods use theoretical knowledge of a domain and a set of classified examples to develop a method for accurately classifying examples not seen during training. The challenge of hybrid learning systems is to use the information provided by one source of information to offset information missing from the other source. By so doing, a hybrid learning system should learn more effectively than systems that use only one of the information sources. KBANN(Knowledge-Based Artificial Neural Networks) is a hybrid learning system built on top of connectionist learning techniques. It maps problem-specific "domain theories", represented in propositional logic, into neural networks and then refines this reformulated knowledge using backpropagation. KBANN is evaluated by extensive empirical tests on two problems from molecular biology. Among other results, these tests show that the networks created by KBANN generalize better than a wide variety of learning systems, as well as several t...

We apply linear genetic programming to several diagnosis problems in medicine. An efficient algorithm is presented that eliminates intron code in linear genetic programs. This results in a significant speedup which is especially interesting when operating with complex datasets as they are occuring in real-world applications like medicine. We compare our results to those obtained with neural networks and argue that genetic programming is able to show similar performance in classification and generalization even within a relatively small number of generations.

We design a molecular Turing machine and determine the complexity of the problems solvable by molecular computers. In [A94], a combinatorial molecular experiment to solve the NP-complete problem of Hamiltonian Path was proposed and implemented. Using our design, we show that such molecular computers can in fact compute PSPACE, under the generous assumptions implicit in [A94]. Under stronger and somewhat more practical restrictions, which [A94] fails to satisfy, we show that molecular computers are limited to solving problems in P. 1 Introduction Molecular and quantum computers have recently been considered as alternatives to standard models such as Turing machines, PRAM's, or logical circuits [S94, Si94, A94, B94]. Shor has shown that quantum computers can factor large numbers and calculate discrete logarithms in polynomial time, even though no known polynomial-time algorithm is known [S94]. Adleman has proposed a molecular "computer" that solves the NPcomplete Hamiltonian Path proble...

Artificial neural networks have proven to be a successful, general method for inductive learning from examples. However, they have not often been viewed in terms of constructive induction. We describe a method for using a knowledgebased neural network of the kind created by the Kbann algorithm as the basis of a system for constructive induction. After training, we extract two types of rules from a network: modified versions of the rules initially provided to the knowledgebased neural network, and rules which describe newly constructed features. Our experiments show that the extracted rules are more accurate, at classifying novel examples, than the trained network from which the rules are extracted. 1 INTRODUCTION Artificial neural networks (ANNs) have proven to be a powerful and general technique for machine learning. For example, a host of empirical comparisons indicate that ANNs are at least as effective at generalizing from training to testing examples as any of several common sym...

this paper we focus on the notion of metabolism and the way it can be expressed in symbolic terms. We claim that an adequate formalization of this concept could help to unify several approaches based on the string rewriting mechanism and could suggest new perspectives in the search for more powerful and general methods in the symbolic description of symbolic dynamical systems [15, 17] where evolution obeys some internal laws of structural recombination. 2 The conceptual development of symbolic systems

this paper was motivated by our previous work on evolutionary design of mechanical conformational switches

We apply an interpreting variant of linear genetic programming to several diagnosis problems in medicine. We compare our results to results obtained with neural networks and argue that genetic programming is able to show similar performances in classi cation and generalization even when using a relatively small number of generations. Finally, an e cient algorithm for the elimination of introns in linear genetic programs is presented.

This paper presents the foundations of p-adic modelling in genomics. Considering nucleotides, codons, DNA and RNA sequences, amino acids and proteins as information systems, we have formulated the corresponding p-adic formalisms for their investigations. Each of these systems has its characteristic prime number used for construction of the related information space. Relevance of this approach is illustrated by some examples. In particular, it is shown that degeneration of the genetic code is a p-adic phenomenon. We have also put a forward a hypothesis on the evolution of the genetic code assuming that primitive code was based on single nucleotides and chronologically first four amino acids. This formalism of p-adic genomic information systems can be implemented in computer programs and applied to various concrete cases.

Abstract not found