Computational Models of Uncertainty Management and Decision Making

一个完整的不确定性管理方法需要支持交互式和增量式的问题表达、推理、假设排序和决策。目前的方法主要集中在推理上，对不确定情况下的推理中的更大问题提供的支持很少，甚至没有。这项提案提出了一项研究计划，研究如何使用多种不确定表示作为资源、有限推理和不确定情况下决策的完整计算基础。研究分为两个阶段。第一阶段是继续开发一种符号/数字相结合的表示法，用于小规模问题模型的制定和评估。该框架提供了支持增量模型构建、评估和修订的基本功能，但没有解决资源约束问题，这些问题在大规模模型的推理中变得至关重要。拟议研究的第二阶段将通过开发将大尺度模型分解成一组小尺度模型的表示和评估策略来解决小尺度模型方法难以解决的问题。

A Complete approach to uncertainty management requires support for interactive and incremental problem formulation, inference, hypothesis ranking, and decision making. Current approaches concentrate primarily on inference, and provide little or no support for the larger issues in reasoning under uncertainty. This proposal presents a plan for research into the use of multiple representations of uncertainty as the basis of a complete computational foundation for resource.bounded reasoning and decision.making under uncertainty. The research is organized into two phases. The first phase is continued development of a combined symbolic/numeric representation for small.scale problem model formulation and evaluation. This framework provides basic capabilities to support incremental model construction, evaluation, and revision, but does not address issues of resource bounding which become critical in reasoning with large.scale models. A second phase of the proposed research will address problems which are intractable for the small.scale model approach, by developing representations and evaluation strategies for decomposing a large.scale model into a set of small.scale models.