Belief formation from observation and belief integration using virtual belief space in Dempster-Shafer probability model

Integrating uncertain information from multiple sources is a key technology to realise reliable AI systems. The Dempster-Shafer probability model (DS model) provides a useful computational scheme for the integration. In this paper, the author proposes two algorithms for belief formation and integration based on the DS model. The first algorithm is for computing a basic probability assignment function based on similarity measures between observed data and object categories. The soundness of the algorithm is shown using mathematical relations between several fuzzy measures. Then, the author proposes a new algorithm for integrating multiple beliefs (i.e, basic probability assignment functions). Using this algorithm, the author can solve a controversial problem in the DS model about how to combine partially conflicting beliefs. That is, with the proposed algorithm, the author can smoothly integrate multiple beliefs even if they are partially/totally conflicting. From a computational viewpoint, moreover, the belief integration by the proposed algorithm can be implemented very efficiently