On a general concept of forgetting

Abstract

Practice leads us to seek a simple method which would make parameter estimation (and subsequent control or signal processing) reliably adaptive. Unfortunately, in most applications we lack sufficient information to specify a complete model of parameter variations. In other words, the problem is ‘under-determined’ which prevents us from employing standard equations of probability calculus. In this paper we apply known principles of rational behaviour in such situations to propose a plausible and well justified solution. The result we get is close to classical exponential forgetting, but regularized by available prior information. We demonstrate the practical implications of this feature.

Notes

‡ SERC Visiting Fellow on leave from the Institute of Information Theory and Automation, Czech Academy of Sciences, P.O. Box 18, 182 08 Prague 8, Czech Republic.