The performance of symbolic inference tasks has long been a challenge to connectionists. In this paper, we present an extended survey of this area. Existing connectionist inference systems are reviewed, with particular reference to how they perform variable binding and rulebased reasoning and whether they involve distributed or localist representations. The benefits and disadvantages of different representations and systems are outlined, and conclusions drawn regarding the capabilities of connectionist inference systems when compared with symbolic inference systems or when used for cognitive modelling.

We introduce an connectionist approach to unification using a local and a distributed representation. A Prolog-System using these unification-strategies has been build. Prolog is a Logic Programming Language which utilizes unification. We introduce a uncertainty measurement in unification. This measurement is based on the structure-abilities of the chosen representations. The strategy using a local representation, called `-CUP, utilizes a self-organizing feature-map (FM-net) to determine similarities between terms and induces the representation for a relaxation-network (relax-net). The strategy using a distributed representation, called d-CUP, embeds a similarity measurement by its recurrent representation. It has the advantage that similar terms have a similar representation. The unification itself is done by a backpropagation network (BP-net). We have proven the systems adequacy for unification, its efficient computation, and the ability to do extended unification. 1 Introduction Un...

this document describes a fully integrated model based on probability theory supported by the method of maximum entropy. It integrates aspects of rule based reasoning and neural networks and its strength lies in well-defined semantics for reasoning with uncertainty, incomplete information, non-monotonic and inductive reasoning.

AppART is an Adaptive Resonance Theory low parameterized neural model that incrementally approximates continuous-valued multidimensional functions from noisy data using biologically plausible processes. AppART performs a higher-order Nadaraya-Watson regression and can be interpreted as a fuzzy logic Standard Additive Model. In this work we describe AppART dynamics and training. We also discuss the approach it makes to hybrid neural systems and deal with its theoretical foundations as a function approximation method. Three benchmark problems are solved in order to study AppART from an application point of view and to compare its results with the ones obtained from other models. Finally two modi cations to the AppART formulation aimed at improving AppART eciency are proposed and tested.

Abstract not found

Abstract not found