Paper Info
Title
Algorithmic Tools for Adversarial Games: Intent Analysis Using Evolutionary Computation.
Abstract
Determining adversarial intent is important on a battlefield. In this paper, we propose a method for intent analysis using evolutionary computation. The proposed approach defines the model as an optimization problem then gives an algorithm for determining the parameters of the model in a Valuated State Space® domain. Example experiments using simulation data are discussed. I. INTRODUCTION One of the keys to successful command and control is to understand the enemy's intent, particularly in light of incomplete and perhaps inaccurate information regarding social and cultural norms. It is inappropriate to project our own goals and aspirations onto the enemy. For example, in asymmetric warfare, the enemy's tactics and objectives may be radically different from our own and those of our allies. Given the same resources to allocate, if placed in obverse roles, our decisions on how best to use those resources might be very different. Furthermore, warfare is becoming increasingly that of semi-autonomous machines versus machines (e.g., swarms of semi-autonomous vehicles reacting to automated defense systems). Understanding the enemy's intent will therefore become less a matter of understanding the thinking of higher command authority and more a matter of inferring the adversary's intent based on a priori beliefs regarding their objectives and observed data reflecting the actual decisions that the enemy takes in real settings. A novel combination of two technologies, evolutionary computation and the Valuated State Space® Approach used to quantify purpose, holds the promise of a general procedure for inferring the enemy's purpose in combat settings ranging from the campaign-level to the level of the individual. The capability described in this report is the result of the research and development undertaken that examined an automatic method for optimizing models of the adversary's intent, structured in a hierarchic form. The models were evolved (optimized) in light of data acquired on decisions made presuming the adversary is rational (i.e., attempting to maximize success as he defines it) using multi-agent adversarial games. The effort developed and tested software to assess the capability of this procedure in simulated combat settings of sufficient complexity. A collection of alternative models of the adversary's purpose was evolved dynamically over time, with evolutionary algorithms used to adapt those models in light of the most recent data describing the observed adversary's behavior. The feasibility of the approach has been assessed in a series of experiments using a statistical design to determine the computational requirements of the procedure and the identifiability of the adversary's objectives as a function of the complexity of the setting. A. Modeling the Enemy as a Problem in System Identification The challenge of inferring the enemy's purpose is similar to the problem of system identification. As indicated in Fig. 1, data are observed regarding the input-output behavior of a system. The goal is to develop a model of the transducer that maps the input stimulus into the output set of observed actions with the least error. The choice of models is often crucial in identifying an appropriate representation of the system. As will be discussed in the next section, the Valuated State Space (VSS) Approach provides the framework for modeling the adversary's mission. Once the class of models is chosen, a search is initiated for the best model of those available. This requires a criterion by which to measure the goodness-of-fit of the model and its associated parameter values to the observed data.
Year
DOI
Venue
2007
10.1109/CISDA.2007.368142
CISDA
Keywords
Field
DocType
evolutionary computation,game theory,military computing,Valuated State Space,adversarial games,algorithmic tools,battlefield,evolutionary computation,intent analysis
Computer science,Theoretical computer science,Artificial intelligence,System identification,Optimization problem,Mathematical optimization,Algorithm design,Battlefield,Computational intelligence,Evolutionary computation,Game theory,Machine learning,Adversarial system
Conference
Citations 
PageRank 
References 
1
0.35
0
Authors
2
Name
Order
Citations
PageRank
Vanesian, T.110.35
Kreutz-Delgado, K.291.11