In this paper we have described the knowledge integration technique and described the results obtained with system INTEG3.1. The results were rather good and show that knowledge integration should be incorporated into ML systems.
Firstly, as we have shown the performance of the integrated theory significantly exceeds the performance of individual theories. In our experiments the performance gains were around 15% which represents a significant improvement.
Also, when learning experiments are repeated with different data, the performance of the integrated theory does not fluctuate as much as the performance of individual theories (standard deviations are relatively small). This is a desirable property. If there are no fluctuations, the theory obtained can be relied upon.
Although size of the integrated theory exceeds somewhat the size of one individual theory, the total number of rules is relatively small. This is an advantage over systems that exploit redundancy [e.g. Gams, 1989].
The authors wish to thank Katarina Morik for useful comments on this work. This work was partly supported by Project Ecoles (Esprit 2 Project 3059). The authors wish to express their gratitude for this support. Gratitute is also expressed to the doctors in Ljubljana and researchers in JSI, Ljubljana who have provided us with the medical data that has been extensively used in our tests.
References
Bond A.H. and Les Gasser (1988): Distributed Artificial Intelligence, Morgan Kaufmann.
Cestnik, B., Kononenko, I. and Bratko, I., (1987): "ASSISTANT 86 : A Knowledge-Elicitation Tool for Sophisticated Users", in Progress in Machine Learning, I.Bratko and N.Lavrac (eds), Sigma Press, Wilmslow, England.
Brazdil, P. and Clark, P. (1990): "Learning from Imperfect Data", in Machine Learning, Meta-Reasoning and Logics, P. Brazdil and K.Konolige (eds.), Kluwer Academic Publishers.
Brazdil, P. and Torgo L. (1990): "Knowledge Acquisition via Knowledge Integration", in Current Trends in Knowledge Acquisition, B. Wielinga et al. (eds.), IOS Press, Amsterdam.
Brazdil, P., Muggleton S. (1990): Learning to Relate Terms in Multi-agent Environment, LIACC, University of Porto. Submitted to EWSL-91.
Cestnik B. (1990): "Estimating Probabilities: A Crucial Task in Machine Learning", in Proc. of ECAI-90, ed. L. Aiello, Pitman Publ.
Clark, P. and Niblett, T. (1987): "Induction in Noisy Domains", in Progress in Machine Learning, I.Bratko and N.Lavrac (eds), Sigma Press, Wilmslow, England.
Clark, P. and Niblett, T. (1989): "The CN2 Induction Algorithm", Machine Learning 3, Kluwer Academic Publishers.
Emde W. (1987): "Non-cumulative Learning in METAXA-3", in Proc. of IJCAI-87, Milan, Italy, pp. 208-212.
Emde W. (1989): "An Inference Engine for Representing Multiple Theories", in Knowledge Representation and Organization in Machine Learning, K. Morik (ed.), pp. 148-176, Springer Verlag.
Gaines B. (1989): Knowledge Acquisition: the Continuum Linking Machine Learning and Expertise Transfer, in Proc of Third European Workshop on Knowledge Acquisition for Knowledge-Based Systems, J.Boose, B.Gaines and J.G.Ganascia (eds), Paris, July 1989.
Gams M. (1989): "New Measurements Highlight the Importance of Redundant Knowledge", in Proceedings of 4th European Working Session on Learning, K. Morik (ed.), pp. 71-80, Pitman - Morgan Kaufmann.
Genesereth, M. and Nilsson, N. (1987): Logical Foundations of Artificial Intelligence, Morgan Kaufmann Publishers Inc.
Good I.J. (1965): The Estimation of Probabilities, M.I.T. Press, Cambridge, Mass.
Janikow C.Z. (1989): "The AQ16 Inductive Learning Program: Some Experimental Results with AQ16 and Other Symbolic and Nonsymbolic Programs", Rep. of AI Center, George Mason University.
Morik K. (1989): "Sloppy Modeling", in Knowledge Representation and Organization in Machine Learning, K. Morik (ed.), pp. 107-134, Springer Verlag.
Murray K.S. and Porter B.W. (1989): "Controlling Search for the Consequences of New Information During Knowledge Integration", in Proc. of 6th International Workshop on Machine Learning, A.M. Segre (ed.), Ithaca, New York, Morgan Kaufmann Inc.
Quinlan J.R. (1986): "The Effect of Noise in Concept Learning", Machine Learning (vol. II), R.Michalski, J.Carbonell and T. Mitchell (eds), Morgan Kaufmann Publishers, Inc.
Shaw M. & Gaines B. (1989): Knowledge Acquisition: Some Foundations, Manual Methods and Future Trends, in Proc. of Third European Workshop on Knowledge Acquisition for Knowledge-Based Systems, J. Boose, B. Gaines and J.G. Ganascia (eds), Paris, July 1989.
Torgo L. (1990): "IRule1: Incremental Rule Learning System", Working Paper, LIACC, University of Porto (in preparation).
Utgoff P.E. (1988): "ID5: An Incremental ID3", in Proc. of 5th International Workshop on Machine Learning, J. Laird (ed.), Ann Harbour, Morgan Kaufmann Inc.
Zhang J. and Michalski R. (1989): "Rule Optimization via SG-TRUNC Method", in Proceedings of 4th European Working Session on Learning, K. Morik (ed.), pp. 251-262, Pitman - Morgan Kaufmann.