Space Efficient Probabilistic Graphical Models and Privacy Sensitive Construction of Agent Organizations

点击翻译按钮获取中文摘要

(1) Rigid, simplistic rules are often used for decision making in personal and mobile devices. For example, a phone number may be placed on a black list, due to the report of a spam call from it, causing future calls from the number to be filtered. The rule ignores the possibility that the report may itself be a spam, leading to undesirable actions. Bayesian Networks (BNs), knowledge based systems capable of weighing complex uncertain context information, can aid users with more intelligent decisions. BNs encode causal relations with graphical structures quantified by probability tables. However, memory needed to run BNs is exponential in the number n of direct causes per variable. When n is large, the required memory may exceed that of a smart phone or sensor, limiting deployment of BNs in such devices.To overcome that, this research seeks to reduce memory requirement for running BNs to being linear in n. It explores innovatively a recent modeling technique, Non-impeding noisy-AND Tree (NAT), to approximate probability tables in BNs. NAT models take the memory linear in n and promise to approximate BN probability tables more accurately than existing techniques. How to best approximate BNs with NAT modeling and how to reason with NAT modeled BNs within linear memory will be investigated. Its success will dramatically reduce memory required for probabilistic reasoning with BNs, allow software engineers to deploy BNs in pervasive computing devices, and enable intelligent decisions in an unprecedented range of applications. (2) Cooperative intelligent systems (called agents) are well suited for applications such as monitoring complex equipment or collaborative design in supply chains. Agent cooperation is often through an organization. The so-called Junction Tree (JT) is one such organization and is found superior than the often used pseudotrees. An agent can embed rich knowledge, e.g., about an equipment component, that is proprietary to component vendor and needs to remain private. However, common methods to construct JT organizations suffer from breach of such privacy. As a result, vendors run the risk of losing intellectual properties.To improve privacy in these intelligent systems, this research studies how to construct JT organizations without privacy loss if possible and with the minimum loss if unavoidable. Flexible JT organization construction will be devised with privacy protection to handle changes in system composition, e.g., when a component and its agent are added. Feasibility of fully autonomous, privacy protecting JT construction, i.e., without using an externally specified leader agent, will be investigated. Successful completion of this research will close a loop hole on privacy in agent systems based on JT organizations. The strong privacy guarantee, coupled with other superior computational properties of JT organizations, will make these agent systems more widely applicable.