Biblio

Attribute reduction algorithms in rough set theory can be classified into two groups, i.e. heuristics algorithms and computational intelligence algorithms. The former has good search efficiency but it can not find the global optimal reduction. Conversely, the latter is possible to find global optimal reduction but usually suffers from premature convergence. To address this problem, this paper proposes a two-stage algorithm for finding high quality reduction. In first stage, a classical differential evolution algorithm is employed to rapidly approach the optimal solution. When the premature convergence is detected, a local search algorithm which is intuitively a forward-backward heuristics is launched to improve the quality of the reduction. Experiments were performed on six UCI data sets and the results show that the proposed algorithm can outperform the existing computational intelligence algorithms.

In this paper, we introduce the concept of transforming attribute-value assignments from one set to another set. We specify two types of transformations–-attribute reduction and attribute expansion. We distinguish policy attributes from non-policy attributes in that policy attributes are used in authorization policies whereas the latter are not. Attribute reduction is a process of contracting a large set of assignments of non-policy attributes into a possibly smaller set of policy attribute-value assignments. This process is useful for abstracting attributes that are too specific for particular types of objects or users, designing modular authorization policies, and modeling hierarchical policies. On the other hand, attribute expansion is a process of performing a large set of attribute-value assignments to users or objects from a possibly smaller set of assignments. We define a language for specifying mapping for the transformation process. We also identify and discuss various issues that stem from the transformation process.