Managing Exponential Decision Spaces (MEDS)
Abstract
Decision-making in the battlefield is based on a huge amount of information that needs to be fused to create not only current situational awareness, but also project the impact to blue forces as well as the potential changes in behavior of red forces. One way to accomplish a set of courses-of-action that will maximize success and satisfy mission objectives is to run a simulation to understand the different potential outcomes and generate probabilities of success based on ouractions. The difficulty of such an approach in the middle of a mission is that you need to run enough simulation runs on an environment that is constantly changing and to attain acceptable confident intervals will be impractical or lack of timeliness. It is important to note that simulations are critical to generate enough heterogeneous scenarios and datasets for training, however there is the need of an alternative to be able to decide best course of action when timeof the essence. Game theory and a number of computerized games (i.e. Chess, StarCraft) are played with well-known moves given certain situational awareness of the current state of the game. Our proposal termed Managing Exponential Decision Spaces (MEDS) will provide set plays (i.e., alternative decision tracks) that are consistent with doctrine. Our proposal will also allow decision-makers to ask what-if questions that will have matched previously analyzed situationsfor which simulation validation have produced a priori results with statistically acceptable confidence. Furthermore, if a decision-maker wants to validate doctrine, the user could ask the system to play the game without any hard constraints. MEDS will assist commanders and decision makers through a variety of analysis and modeling techniques that automate the evaluation of options to take at any given state while presenting the best alternatives in a clear and concise manner for final approval. The key aspect of our proposed work is to manage a Decision-Space that could grow exponentially. Therefore, the main technical challenge is to balance a scalable Decision Tree against maintaining the most plausible/impactful COAs and ECOAs over the life of the mission. With these objectives in mind, MEDS will have three main components: AI/ML Fusion and Relevant Events Engine: Once a Mission is defined, a composition of feasible Blue Tasks (BTs), predicted/estimated possible Red Tasks (RTs) and Measure of Effectiveness (MoEs) are fused in creating atomic cubes that are defined as triplets. Each triplet is (BTi, RTj, MoEk). This will create a 3-dimentional space with axis been the BTs, RTs and MoEs (shown in Figure 1 as the small cubes forming the overall Decision Space Cube) AI/ML to Generate and Manage Decision Space: Once the atomic elements (small cube) is created and the overall Decision Space Cube (made of all the atomic cubes in the 3- dimensional space), then we need to define decision paths that are represented as a DecisionTree or Forest. AI/ML to Cluster and Discover Decision Tracks: Once we have manageable size Decision Tree or Forest, we need a service that will create and traverse Decision Tracks. Given that there could be exponential number of such tracks (even in a truncated Decision Space) the AI/ML tools will have the biggest potential benefit in recommending the most effective decision tracks to follow to satisfy the mission objective.Managing Exponential Decision Spaces (MEDS)
Document Details
- Document Type
- DoD Grant Award
- Publication Date
- Apr 29, 2020
- Source ID
- N000142012242
Entities
People
- Mosies Sudit
Organizations
- Office of Naval Research
- Research Foundation for the State University of New York
- United States Navy