No description available.

No description available.

An issue with learning effective policies in multi-agent adversarial games is that the size of the search space can be prohibitively large when the actions of both teammates and opponents are considered simultaneously. Opponent modeling, predicting an opponent’s actions in advance of execution, is one approach for selecting actions in adversarial settings, but it is often performed in an ad hoc way. In this dissertation, we introduce several methods for using opponent modeling, in the form of predictions about the players’ physical movements, to learn team policies. To explore the problem of decision-making in multi-agent adversarial scenarios, we use our approach for both offline play generation and real-time team response in the Rush 2008 American football simulator. Simultaneously predicting the movement trajectories, future reward, and play strategies of multiple players in real-time is a daunting task but we illustrate how it is possible to divide and conquer this problem with an assortment of data-driven models. By leveraging spatio-temporal traces of player movements, we learn discriminative models of defensive play for opponent modeling. With the reward information from previous play matchups, we use a modified version of UCT (Upper Conference Bounds applied to Trees) to create new offensive plays and to learn play repairs to counter predicted opponent actions. iii In team games, players must coordinate effectively to accomplish tasks while foiling their opponents either in a preplanned or emergent manner. An effective team policy must generate the necessary coordination, yet considering all possibilities for creating coordinating subgroups is computationally infeasible. Automatically identifying and preserving the coordination between key subgroups of teammates can make search more productive by pruning policies that disrupt these relationships. We demonstrate that combining opponent modeling with automatic subgroup identification can be used to create team policies with a higher average yardage than either the baseline game or domain-specific heuristics.

This dissertation reports on an effort to design, construct, test, and adjust an expert system for making certain business decisions. A widely used approach to recurring judgmental decisions in business and other social organizations is the "rule-based decision system". This arrangement employs staff experts to propose decision choices and selections to a decisionmaker. Such decisions can be very important because of the large resources involved. Rules and values encountered in such systems are often vague and uncertain. Major questions explored by this experimental effort were: (1) could the output of such a decision system be mimicked easily by a mechanism incorporating the rules people say they use, and (2) could the imprecision endemic in such a system be represented by fuzzy set constructs. The task environment chosen for the effort was a computer-based game which required player teams to make a number of interrelated, recurring decisions in a realistic business situation. The primary purpose of this research is to determine the feasibility of using these methods in real decision systems. The expert system which resulted is a relatively complicated, feed-forward network of "simple" inferences, each with no more than one consequent and one or two antecedents. Rules elicited from an expert in the game or from published game instructions become the causal implications in these inferences. Fuzzy relations are used to represent imprecise rules and two distinctly different fuzzy set formats are employed to represent imprecise values. Once imprecision appears from the environment or rules the mechanism propagates it coherently through the inference network to the proposed decision values. The mechanism performs as well as the average human team, even though the strategy is relatively simple and the inferences crude linear approximations. Key aspects of this model, distinct from previous work, include: (1) the use of a mechanism to propose decisions in situations usually considered ill-structured; (2) the use of continuous rather than two-valued variables and functions; (3) the large scale employment of fuzzy set constructs to represent imprecision; and (4) use of feed forward network structure and simple inferences to propose human-like decisions.