Loading...
Please wait, while we are loading the content...
Similar Documents
Interactive Planning under Uncertainty with Causal Modeling and Analysis
| Content Provider | Semantic Scholar |
|---|---|
| Author | Kuter, Ugur |
| Copyright Year | 2003 |
| Abstract | This paper describes a new technique for interactive planning under conditions of uncertainty. Our approach is based on the use of the Air Force Research Laboratory’s Causal Analysis Tool (CAT), a system for creating and analyzing causal models similar to Bayes networks. In order to use CAT as a tool for planning, users go through an iterative process in which they use CAT to create and analyze alternative plans. One of the biggest difficulties is that the number of possible plans is exponential. In any planning problem of significant size, it is impossible for the user to create and analyze every possible plan; thus users can spend days arguing about which actions to include in their plans. To solve this problem, we have developed a way to quickly compute the minimum and maximum probabilities of success associated with a partial plan, and use these probabilities to recommend which actions the user should include in the plan in order to get the plan that has the highest probability of success. This provides an exponential reduction in amount of time needed to find the best plan. Problem and Significance A major feature of military plans is the huge amount of uncertainty they contain. This uncertainty is often referred to as the ”fog of war.” There are many sources for this uncertainty; perhaps the major source of this uncertainty is the relationship between cause and effect. For example: At a tactical level, sorties are flow against a series of bridges to prevent the enemy ground forces from crossing the river. The sorties are intended to prevent the crossing. What is the probability that they will? At a strategic level, destruction of the Taliban Army was intended ultimately to reduce worldwide terrorism. Did it? Due to this uncertainty, the number of possible plans for carrying out a military operation successfully can be quite large. Quick and accurate decision making on which of the series of actions to take is a very important task and it is very hard. The decision making process highly relies on the information gathered about those factors, which is often uncomplete. A typical military plan involves many such sources of uncertainty. For example, a causal model of Operation Deny Freedom, built by the actual planners, contains over 300 uncertain events interrelated by cause and effect. Moreover, there are often significant delays between cause and effect, and effects may persist for only limited amounts of time: a bridge destroyed by air power can be rebuilt or bypassed. This paper describes the tool that we are developing to help manage this uncertainty in order to develop effective plans. The basis for our approach is the Air Force Research Laboratory’s (AFRL’s) Causal Analysis Tool (CAT), which is a tool for representing and analyzing causal networks similar to Bayesian networks. In order to represent plans using CAT’s causal networks, all of the actionable items, i.e., the actions that might potentially appear in a plan, are represented as nodes within the causal network. Thus, each possible combination of actionable items is a possible plan. From this representation, CAT can compute the probability that any given plan (i.e., any chosen combination of actionable items) will achieve the desired objectives. A major technical difficulty is how to overcome combinatorial blowup during the planning process. If there are n different actionable items, then there are potentially 2 different plans, making it infeasible for the user to ask CAT to analyze each one. One result of this problem is that users of CAT can spend days arguing about which subsets of actionable items to use as their plans. As described in this paper, we have developed a new approach for overcoming this combinatorial blowup. Our approach exploits conditional independence within the causal network in order to compute quick and accurate feedback to the user about how the best way to extend a partial plan into a complete plan. We summarize the theory underlying our approach, describe how we have implemented it by modifying CAT, and give examples of its operation in order to demonstrate the effectiveness of our approach. |
| File Format | PDF HTM / HTML |
| Alternate Webpage(s) | http://drum.lib.umd.edu/bitstream/handle/1903/1251/CS-TR-4434.pdf?isAllowed=y&sequence=1 |
| Alternate Webpage(s) | http://www.dtic.mil/dtic/tr/fulltext/u2/a447944.pdf |
| Alternate Webpage(s) | http://www.cs.umd.edu/Library/TRs/CS-TR-4434/CS-TR-4434.pdf |
| Alternate Webpage(s) | https://www.cs.umd.edu/Library/TRs/CS-TR-4434/CS-TR-4434.pdf |
| Language | English |
| Access Restriction | Open |
| Subject Keyword | Bayesian network Borg Scale Rating of Perceived Exertion Score 17 Causal filter Causal model Causality Decision Making Exclusion HL7PublishingSubSection |
| Content Type | Text |
| Resource Type | Article |
