Applying intent-sensitive policy to automated resource allocation: command, communication and most importantly, control

msra(2000)

引用 27|浏览13
暂无评分
摘要
As battlefield communications technologies have begun to achieve their potential, the notion of ‘winning the information war’ has begun to take on a new complexion—one in which we may be our own biggest enemy. A fundamental problem is that the overhead costs and complexity associated with deciding how to allocate information resources—including generation, processing and routing resources—are too high in an environment where human decision and guidance resources are already overstressed. We are developing an approach that enables commanders to specify policy that will inform an automated communications resource management system. This policy represents the commander’s intent for the allocation of communications resources during the execution of a mission; the policy takes the form of a set of general and specific statements about the priorities, constraints and objectives for information flow. We view this policy as central to any problem in which a decision-maker wishes to precisely guide the behavior of a resource allocation actor. As such, policy has utility in complex domains ranging from cockpit display space to refinery operations. Our approach is called IPSO-FACTO—Intuitive Policy Specification for Optimized Flow of Asynchronous CI Transmissions in Operations. IPSO-FACTO addresses problems including: suitable representations of policy, policy conformance metrics, adaptive information allocation, multi-user policies, and semi-automated policy construction. We have implemented a prototype version of the IPSOFACTO system (that does not yet incorporate this taskbased policy derivation) and are currently evaluating it. Initial results show that the policy representation we have provided is very expressive, but time consuming and error prone to generate directly. Nevertheless, policy created in such a fashion does enable human control of automated resource allocation algorithms and can improve the performance of such resource allocation dramatically. 1 This work was performed under funding from DARPA contract DABT63-99C-0003 INTRODUCTION As sensor and processor capabilities continue to increase, the amount of current, potentially relevant information continues to exceed the available communications bandwidth. Even when a user is dedicated to the task of deciding what information is worthy of transmission (e.g., when surfing the web), the analysis task is simply too big, and the results are often not quite what was intended. In an environment where human decision and guidance resources are already overstressed, the overhead costs and complexity associated with deciding how to allocate information resources are simply unaffordable. While this problem exists for most actors on the battlefield, it is most critical for the tactical commander. The commander could delegate this task to subordinates or automation, but this invites mismatches between the commander’s goals and intentions and the information policies that are enforced. Intelligent systems are being developed to better manage communication networks, but how can a commander successfully convey his intentions to one of these complex systems, so that the crucial information gets to the right soldier at the right time? In our view, the solution must comprise the following components: Policy Representation—a syntactic formulation that balances the need for expressivity and comprehensibility, Policy Conformance Metr ics—a computational framework that allows evaluation of a given proposed solution against the expressed policy, Adaptive Information Control (AIC)—a resource allocation mechanism that is sensitive to the interaction of expressed policy with world state, Multi-user Policies —a method to allow multiple users with differing interests and scopes of control authority to work collaboratively to establish a single effective policy, Semi-automated Policy Elaboration and Conversion (SPEC)—a means to assist a user in specifying what may be a fairly detailed policy in a rapid, intuitive fashion.
更多
查看译文
AI 理解论文
溯源树
样例
生成溯源树,研究论文发展脉络
Chat Paper
正在生成论文摘要