Concept support as a method for programming neural networks with symbolic knowledge
Neural networks are usually seen as obtaining all their knowledge through training on the basis of examples. In many AI applications appropriate for neural networks, however, symbolic knowledge does exist which describes a large number of cases relatively well, or at least contributes to partial solutions. From a practical point of view it appears to be a waste of resources to give up this knowledge altogether by training a network from scratch. This paper introduces a method for inserting symbolic knowledge into a neural network-called “concept support.” This method is non-intrusive in that it does not rely on immediately setting any internal variable, such as weights. Instead, knowledge is inserted through pre-training on concepts or rules believed to be essential for the task. Thus the knowledge actually accessible for the neural network remains distributed or -as it is called-subsymbolic. Results from a test application are reported which show considerable improvements in generalization.
KeywordsNeural Networks Connectionism Preprogramming Application
Unable to display preview. Download preview PDF.
- [Gallant 88]
- [vanGelder 90]Tim van Gelder 1990. Why Distributed Representation is Inherently Non-Symbolic. In: Georg Dorffner (Ed.), Konnektionismus in Artificial Intelligence und Kognitionsforschung, Proc.of the 6th Austrian AI-Conference, Springer Berlin 1990, pp. 58–66.Google Scholar
- [Hendler 89]J.A. Hendler 1989. On the Need for Hybrid Systems. Connection Science 3(1), pp. 227–229.Google Scholar
- [Jacobs 88]Robert A. Jacobs 1988. Initial Experiments On Constructing Domains of Expertise and Hierachiers in Connectionist Systems. In: D.S. Touretky (Ed.) Proc.of the Connectionist Models Summer School 1988, Morgan Kaufman, San Mateo, CA, pp. 144–153.Google Scholar
- [Lendaris 90]George C.Lendaris, Ihab A.Harb 1990. Improved Generalization in ANN's via Use of Conceptual Graphs: A Character Recognition Task as an Example Case. In: Proc.IJCNN 1990, San Diego, CA, 1990, pp.I 551–555.Google Scholar
- [Porenta 88]Gerold Porenta, G. Dorffner, J. Schedlmayer, H. Sochor 1988. Parallel Distributed Processing as a Decision Support Approach in the Analysis of Thallium-201 Scintigrams. In: Proc. Computers in Cardiology 1988, Washington D.C., IEEE.Google Scholar
- [Sethi 90]
- [Suddarth 90]S.C. Suddarth, Y.L. Kergosien 1990. Rule-injection hints as a means of improving network performance and learning time. In: L.B. Almeida, C.J. Wellekens (Eds.), Neural Networks, Proc. of the 1990 EURASIP Workshop, Springer, Berlin, 1990.Google Scholar
- [Yang 90]Qing Yang, Vijay K.Bhargava 1990. Building Expert Systems by a Modified Perceptron Network with Rule-Transfer Algorithms. Proc.of the IJCNN 1990, San Diego, CA, pp.II 77–82.Google Scholar
- [Yeoung-Ho 90]Yu Yeoung-Ho, Robert F.Simmons 1990. Extra Output Biased Learning. Proc.of the 1990 IJCNN, San Diego, CA, 1990, pp.III 161–164.Google Scholar
- [Wiklicky 90]Herbert Wiklicky 1990. Neurale Strukturen: Möglichkeiten der Initialisierung und symbolisches Wissen. Thesis, University of Vienna, 1990.Google Scholar