par Nguyen, Hung;Smets, Philippe
Référence International journal of approximate reasoning, 8, 2, page (89-104)
Publication Publié, 1993-02
Article révisé par les pairs
Résumé : The concept of knowledge or belief so important in Artificial Intelligence (AI) has been discussed in depth by philosophers, exemplified by Gärdenfors [1]. Whereas it is useful to have a broad view on the topic, it is also necessary to look at concrete models (eg, numerical modeling of degrees of belief). In this paper, we will discuss the cautious aspect of Shafer's approach to belief modeling and investigate conditionals in the context of belief dynamics. In particular, we show that Dempster's rule of conditioning produces the least committed belief function among the set of belief functions generated by conditional objects, thus providing a justification for its use. © 1993.