A Reflective Architecture for
Integrated
Memory-based Learning and Reasoning

	Josep Llus Arcos	Enric Plaza


Artificial Intelligence Research Institute, IIIA.
Spanish Council for Scientific Research, CSIC.
Cam de Santa Brbara, 17300 Blanes, Catalunya, Spain.
{arcos | plaza}@ceab.es



Abstract. In this paper we will discuss the role of case-based reasoning and
learning as a tool for integrating different methods of inference and different
methods of learning. The Massive Memory Architecture, an experimental
framework for experience-based learning and reasoning, is described. Its
reflective capabilities are described and we put forth the hypothesis that learning
methods are inference methods able to inspect the problem solving process and
modify the system itself so as to improve its behavior. Therefore, learning
methods require a self-model of the system. Self-models and method
implementation are based on conceptual, knowledge-level descriptions of
inference.
References
[1] Aamodt, A., Knowledge-intensivecase-based reasoning and learning. Proc. ECAI-
90, Stockholm, August 1990.
[2] Akkermans, H., van Harmelen, F., Schreiber, G., Wielinga, B., A formalisation of
knowledge-level model for knowledge acquisition. Int Journal of Intelligent Systems
forthcoming, 1993.
[3] Armengol, E. and Plaza E., Analyzing case-based reasoning at the knowledge
level. European Workshop on Case-based Reasoning EWCBR 93, 1994.
[4] Carbonell, J., Derivational analogy: A theory of reconstructive problem solving
and expertise acquisition. In R. S. Michalski, J. G. Carbonell, and T. M. Mitchell
(Eds.),Machine Learning , Vol. II. Morgan Kaufmann, pp.371-392, 1986.

[5] Carbonell, J. G., Knoblock, C. A., Minton, S., Prodigy: An integrated architecture
for planning and learning. In K Van Kehn (Eds.), Architectures for Intelligence.
Lawrence Erlbaum Ass., Hillsdadale, NJ, 1991.

[6] Chandrasekaran, B., Task structures, knowledge acquisition and machine learning.
Machine Learning 2:341-347, 1989.

[7] Giunchilia, F., and Traverso, P., Plan formation and execution in an architecture of
declarative metatheories . Proc of META -90: 2nd Workshop of Metaprogramming in
Logic Programming.. MIT Press, 1990.

[8] Godo, L., L6pez de Mntaras, R., Sierra, C., Verdaguer, A., MILORD: The
architecture and the management of linguistically expressed uncertainty. Int. J.
Intelligent Systems, 4:47 1-501, 1989.

[9] Greiner, R., Lenat, D. RLL- 1: A Representation Language Language, HPP-80-9
Comp. Science dept., Stanford University. Expanded version of the same paper in
Proc. First AAAI Conference., 1980.

[10] Kiczales G., Des Rivires J., Bobrow D. G., The Art of the Metaobject Protocol,
The MIT Press: Cambridge, 1991.

[11] Lpez, B. and Plaza, E., Case-based planning for medical diagnosis, In Z Ras
(Ed.) Methodologies for Intelligent Systems. Lecture Notes in Artificial Intelligence,
689, p. 96-105. Springer-Verlag, 1993.

[121 Mitchell, T.M., Allen, J., Chalasani, P., Cheng, J., Etzioni, 0., Ringuette, M.,
Schlimmer, J. C. , Theo: a framework for self-improving systems. In K Van Lenhn
(Ed.) Architectures for Intelligence. Laurence Erlbaum, 1991.

[13] Newell, A., Unified Theories of Cognition. Cambridge MA: Harvard
UniversityPress, 1990.
[14] Plaza, E, Reflection for analogy: Inference-level reflection in an architecture for
analogical reasoning. Proc. IMSA 92 Workshop on Reflection and Metalevel
Architectures, Tokyo, p. 166-171, November 1992.
[15] Plaza, E. Arcos J. L., Reflection and Analogy in Memory-based Learning, Proc.
Multistrategy Learning Workshop. p. 42-49, 1993.

[16] Plaza, E. Arcos J. L., Flexible Integration of Multiple Learning Methods into a
Problem Solving Architecture, Reserch Report 93/16 1993.

[17] Ram, A., Cox, M .T., Narayanan, S., An architecture for integrated introspective
learning. Proc. ML 92 Workshop on Computational Architectures for Machine
Learning and Knowledge Acquisition, 1992.

[18] Russell, S., The use of knowledge in analogy and induction. Morgan Kaufinann,
1990.

[19] Sierra, C., and Godo, L. Specifying simple scheduling tasks in a reflective and
modular architecture. In J Treur and T Wetter (Eds.) Formal Specifications Methods
for Complex Reasoning Systems,.Ellis Horwood, pp. 199-232, 1993.

[20] Slodzian, A., Configuring decision tree learning algorithms with KresT,
Knowledge level models of machine learning Workshop preprints. Catania, Italy,
April 1994.

[21] Smith, B. C., Reflection and semantics in a procedural language, In Brachman, R.
J., and Levesque, H. J. (Eds.) Readings in Knowledge Representation. Morgan
Kauffman, California, pp. 31-40, 1985.

[22] Steels, L., The Components of Expertise, Al Magazine, Summer 1990.

[23] Treur, J., On the use of reflection principles in modelling complex reasoning. Int.
J. Intelligent Systems , 6:277-294, 1991.

[24] Van de Velde, W., Towards Knowledge Level Models of Learning Systems,
Knowledge level models of machine learning Workshop preprints. Catania, Italy,
April 1994.

[25] van Marcke, K., KRS: An object-oriented representation language, Revue
dIntelligence Artificielle, 1(4), 43-68, 1987.

[26] Wielinga, B., Schreiber, A., Breuker, J., KADS: A modelling approach to
knowledge engineering. Knowledge Acquisition 4(1), 1992.

[27] Wielinga, B., Van de Velde, W., Schreiber, G., Akkermans, H., Towards a
Unification of Knowledge Modelling Approaches. In J.-M. David, J.-P. Krivine and
R. Simmons (eds.). Second generation Expert Systems. pp299-335. Springer Verlag,
1993.
