This paper considers the task of predicting the future decisions of an agent A based on his past decisions. We assume that A is rational - he uses the principle of maximum expected utility. We also assume that the probability distribution P he assigns to random events is known, so that we need only infer his utility function u to model his decision process. We consider the task of using A's previous decisions to learn about u. In particular, A's past decisions can be viewed as constraints on u. If we have a prior probability distribution p(u) over u (e.g., learned from a set of utility functions in the population), we can then condition on these constraints to obtain a posterior distribution q(u). We present an efficient Markov Chain Monte Carlo scheme to generate samples from q(u), which can be used to estimate not only a single "expected" course of action for A, but a distribution over possible courses of action. We show that this capability is particularly useful in a two-player setting where a second learning agent is trying to optimize her own payoff, which also depends on A's actions and utilities.

This article examines prospects for theories and methods of preferences, both in the specific sense of the preferences of the ideal rational agents considered in economics and decision theory and in the broader interplay between reasoning and rationality considered in philosophy, psychology, and artificial intelligence. Modern applications seek to employ preferences as means for specifying, designing, and controlling rational behaviors as well as descriptive means for understanding behaviors. We seek to understand the nature and representation of preferences by examining the roles, origins, meaning, structure, evolution, and application of preferences.

Decision-theoretic optimization is becoming a popular tool in the user interface community, but creating accurate cost (or utility) functions has become a bottleneck --- in most cases the numerous parameters of these functions are chosen manually, which is a tedious and error-prone process. This paper describes ARNAULD, a general interactive tool for eliciting user preferences concerning concrete outcomes and using this feedback to automatically learn a factored cost function. We empirically evaluate our machine learning algorithm and two automatic query generation approaches and report on an informal user study.

Innovations in wireless sensor networks (WSNs) have dramatically expanded the applicability of control technology in day-to-day life, by enabling the cost-effective deployment of large scale sensor-actuator systems. In this paper, we discuss the issues and challenges involved in deploying control-oriented applications over unreliable, resource-constrained WSNs, and describe the design of our planned Sensor Control System (SCS) that can enable the rapid development and deployment of such applications. 1

Intelligent agents often need to assess user utility functions in order to make decisions on their behalf, or predict their behavior. When uncertainty exists over the precise nature of this utility function, one can model this uncertainty using a distribution over utility functions. This view lies at the core of games with incomplete information and, more recently, several proposals for incremental preference elicitation. In such cases, decisions (or predicted behavior) are based on computing the expected expected utility (EEU) of decisions with respect to the distribution over utility functions. Unfortunately, decisions made under EEU are sensitive to the precise representation of the utility function. We examine the conditions under which EEU provides for sensible decisions by appeal to the foundational axioms of decision theory. We also discuss the impact these conditions have on the enterprise of preference elicitation more broadly.

Abstract not found

This paper presents an agent strategy for complex bilateral negotiations over many issues with inter-dependent valuations. We use ideas inspired by graph theory and probabilistic influence networks to derive efficient heuristics for negotiations about multiple issues. Experimental results show — under relatively weak assumptions with respect to the structure of the utility functions – that the developed approach leads to Pareto-efficient outcomes. Moreover, Pareto-efficiency can be reached with few negotiation steps, because we explicitly model and utilize the underlying graphical structure of complex utility functions. Consequently, our approach is applicable to domains where reaching an efficient outcome in a limited amount of time is important. Furthermore, unlike other solutions for high-dimensional negotiations, the proposed approach does not require a mediator.

Non-expected utility (non-EU) theories, such as rank dependent utility (RDU) theory, have been proposed as alternative models to EU theory in decision making under risk. These models do not share the separability property of EU theory hence, in dynamic decision problems, the sophisticated strategy is likely to be dominated w.r.t. stochastic dominance. Although a rational non-EU behavior is necessarily a non-consequentialist behavior (i.e., the choice in a subtree depends on what happens in the rest of the tree), we show that it is nonetheless possible to dene a procedure which: i) involves a "rolling back" of the decision tree; and ii) selects a non-dominated strategy that realizes a compromise between the decision maker's discordant goals at the dierent decision nodes. Relative to the computations involved in the standard EU evaluation of a decision problem, the main computational increase is due to the identication of non-dominated strategies by linear programming. A simulation, using the RDU criterion, conrms the computational tractability of the model. 1

Abstract. Frameworks for handling decision problems have been subject to many advances in the last years, both w.r.t. representation languages, solution algorithms and methods for analyzing decision problems. In this paper we outline some of the recent advances by taking outset in the influence diagram framework. In particular, we shall focus on advances in representation languages and exact solution algorithms for decision problems with a single decision maker. Moreover, we give a brief outline of recent contributions to methods for performing sensitivity analysis in influence diagrams. 1

Multiagent resource allocation is a timely and exciting area