A Utility-Based Approach to Learning in a Mixed
Case-Based and Model-Based Reasoning Architecture


Maarten van Someren1, Jerzy Surma2 and Pietro Torasso3

1 Universiteit van Amsterdam

Roetersstraat 15
1018 WB Amsterdam
The Netherlands
email: maarten@swi.psy.uva.nl

2 Limhurg University Centre

Universitaire Campus D
B-3590 Diepenbeek
Belgium
email: surma@rsftew.luc.ac.be

3 Dipartimento di Informatica
Universita di Torino
Corso Svizzera 185
Torino
Italia
email: torasso@di.unito.it


Abstract. Case-based reasoning (CBR) can be used as a form of caching
solved problems to speedup later problem solving. Using cached cases
brings additional costs with it due to retrieval time, case adaptation time
and also storage space. Simply storing all cases will result in a situation
in which retrieving and trying to adapt old cases will take more time
(on average) than not caching at all. This means that caching must be
applied selectively to build a case memory that is actually useful. This is
a form of the utility problem [4, 2]. The approach taken here is to construct 
a cost model of a system that can be used to predict the effect
of changes to the system. In this paper we describe the utility problem
associated with caching cases and the construction of a cost model.
We present experimental results that demonstrate that the model can
be used to predict the effect of certain changes to the case memory.
References

[1]	Console, L., Portinale, L., Theseider, D., and Torasso, P. (1993). Combining Heuristic 
Reasoning with Causal Reasoning In Diagnostic Problem Solving, pages 4668.
Springer Verlag.
[2]	Francis, A. G. and Ram, A. (1995). A comparative utility analysis of case-based
reasoning and controle-rule learning systems. In Lavrac, N. and Wrobel, S., editors,
Machine Learning: ECML-95, pages 138150. Springer Verlag.
[3]	Keller, R. M. (1988). Defining operationality for explanation-based learning. Artificial 
Intelligence, 35:227241.
[4]	Minton, 5. (1988). Learning Effective Search Control Knowledge: An Explanation-Based 
Approach. Kluwer.
[5]	Portinale, L. and Torasso, P. (1995). Adapter: an integrated diagnostic system
combining case-based and abductive reasoning. In Veloso, M. and Aamodt, A.,
editors, Proceedings ICCBR-95, pages 277288. Springer Verlag.
[6]	Portinale, L. and Torasso, P. (1996). On the usefulness of re-using diagnostic solutions. 
In Wahlster, W., editor, Proceedings 12th European Conference on Artificial
Intelligence ECAI-96, pages 137141. John Wiley and Sons.
[7]	Smyth, B. and Keane, M. (1995). Remembering to forget. In Mellish, C., editor,
Proceedings IJCAI-95, pages 377382. Morgan Kaufmann.
[8]	Straatman, R. and Beys, P. (1995). A performance model for knowledge-based systems. 
In Ayel, M. and Rousse, M. C., editors, EUROVAV-95 European Symposium
on the Validation and Verification of Knowledge Based Systems, pages 253263.
ADEIRAS, Uuiversit de Savoie, Chambry.
[9]	Subramanian, D. and Hunter, 5. (1992). Measuring utility and the design of provably 
good ebi algorithms. In Sleemau, D. and Edwards, P., editors, Machine Learning: 
Proceedings of the Ninth International Workshop ML-95, pages 426435. Morgan Kaufmann.
[10] van Harmelen, F. (1994). A model of costs and benefits of meta-level computation.
In Proceedings of the Fourth Workshop on Meta-programming in Logic (ME TA 94),
volume 883 of LNCS, pages 248261. Springer-Verlag.
[11] van Someren, M. W., Zheng, L. L., and Post, W. (1990). Cases, models or compiled 
knowledge? - a comparative analysis and proposed integration. In Wielinga,
B. J., Boose, J., Gaines, B., Schreiber, G., and van Someren, M. W., editors, Current
trends in knowledge acquisition, pages 339355. lOS Press.
