see my ceevee

if you have trouble downloading a paper or for more information contact


"Measuring the Resilience of Advanced Life Support Systems"
with J. Levri and R. Dearden

Despite the central importance of crew safety in the design and operation of a life support system, the tools commonly used to evaluate alternative Advanced Life Support (ALS) technologies do not currently provide explicit techniques for measuring safety. The resilience of a system, or the system’s ability to meet performance requirements and recover from component-level faults, is a fundamentally dynamic property. This paper motivates the use of computer models as a tool to understand and improve system resilience throughout the design process. Extensive simulation of a hybrid computational model of a water revitalization subsystem (WRS) with probabilistic, component-level faults provides data about off-nominal behavior of the system. The data are used to consider alternative measures of resilience as predictors of the system’s ability to recover from component-level faults. A novel approach to measuring system resilience using a Markov chain model of performance data is also developed. Results emphasize that resilience depends on the complex interaction of faults, controls, and system dynamics, rather than on simple fault probabilities.


"Coordination Failure as a Source of Congestion"
with W. A. Sethares, and J. A. Bucklew
forthcoming in IEEE Transactions on Signal Processing.

Coordination failure, or agents' uncertainty about the action of other agents, may be an important source of congestion in large decentralized systems. The El Farol problem provides a simple paradigm for congestion and coordination problems that may arise with over utilization of the Internet. This paper reviews the El Farol problem and surveys previous approaches, which typically involve complex deterministic learning algorithms that exhibit chaotic-like trajectories. This paper recasts the problem in a stochastic framework and derives a simple adaptive strategy that has intriguing optimization properties; a large collection of decentralized decision makers, each acting in their own best interests and with limited knowledge, converge to a solution that (optimally) solves a complex congestion and social coordination problem. A variation in which agents are allowed access to full information is not nearly as successful. The algorithm, which can be viewed as a kind of habit formation, is analyzed using a weak convergence approach, and simulations illustrate the major results.


"Taking Externalities Seriously: An Economic Perspectice on the Precautionary Principle"
(alternative download site)
Redefining Progress

Summary also available in html or pdf format

The mounting threats to human health and the environment demand that we start taking externalities seriously. It's time to rethink our approach to managing the potential side effects of new technologies and industrial production. The precautionary principle looks at the problem of externalities from a common sense, prevention-oriented viewpoint. It considers all of the different facets of a problem and searches for the alternative with the fewest costly externalities. Rather than asking how much toxic pollution we can live with, the precautionary principle asks what kind of a world do we want to live in, and provides a decision making framework for getting us there.


"Avoiding Global Congestion Using Decentralized Adaptive Agents"
with W. A. Sethares
IEEE Transactions on Signal Processing, Vol. 49, No. 11, November 2001.

Everyone wants to go to a bar called El Farol if it is not crowded but would rather stay home if it is. Unfortunately, the only way to know whether or not the bar is crowded is to go. While such a scenario appears far removed from the typical communications literature, it provides a simple paradigm for analyzing public goods like the Internet, which may simultaneously suffer from congestion and coordination problems, e.g., multiple users trying to connect to the same server or to use the same resource simultaneously. This paper reviews previous solutions to the El Farol Santa Fe bar problem, which typically involve complex learning algorithms. A simple adaptive strategy similar to many signal processing algorithms such as LMS and its signed variants is proposed. The strategy is investigated via simulation, and the algorithm is analyzed in a few simple cases. Unlike most signal processing applications, the objective of the adaptation is not fast and accurate parameter estimation but rather the achievment of a degree of global coordination among users.

Index Terms—Coordination failure, El Farol, multiagent systems, network congestion.


"Dynamically Interdependent Preferences in a General Equilibrium Environment"
Journal of Economic Behavior and Organization, Vol. 47 309-333 2002.

This paper explores the consequences of interdependent preferences for consumer goods, that is, preferences that evolve in response to the consumption decisions of neighboring agents. The key feature is that the interdependence of preferences coexists and interacts with the price mechanism in a general equilibrium environment. The interaction between the negative feedback operating through the price system and the positive feedback expressed in the bandwagon effect creates distinct geographic patterns of consumption on the micro-level and a characteristic evolution of average preferences and production on the macro-level. In equilibrium, agents' preferences and consumption are completely polarized into stable regions in which every agent consumes the same good exclusively.


"Reinforcement Learning in a Nonstationary Environment" (1.8 MB)
Computational Economics, Vol. 18, 89-111, 2001.

This paper examines the performance of simple learning rules in a complex adaptive system based on a coordination problem modeled on the {\em El Farol} problem. The key features of the {\em El Farol} problem are that it typically involves a medium number of agents and that agents' payoff functions have a discontinuous response to increased congestion. First we consider a single adaptive agent facing a stationary environment. We demonstrate that the simple learning rules proposed by Roth and Er'ev can be extremely sensitive to small changes in the initial conditions and that events early in a simulation can affect the performance of the rule over a relatively long time horizon. In contrast, a reinforcement learning rule based on standard practice in the computer science literature converges rapidly and robustly. The situation is reversed when multiple adaptive agents interact: the RE algorithms often converge rapidly to a stable average aggregate attendance despite the slow and erratic behavior of individual learners, while the CS based learners frequently over-attend in the early and intermediate terms. The symmetric mixed strategy equilibria is unstable: all three learning rules ultimately tend towards pure strategies or stabilize in the medium term at non-equilibrium probabilities of attendance. The brittleness of the algorithms in different contexts emphasize the importance of thorough and thoughtful examination of simulation-based results.


"Advanced Control Techniques For Efficient And Robust Operation Of Advanced Life Support Systems"
with C.W. Pawlowski, S. Crawford, W. A. Sethares, and C. Finn
31st International Conference on Environmental Systems, Orlando FL, July 2001. Paper number 01-ICES-208.

This paper examines the structure and performance of three control strategies for a regenerative life support system constrained by mass balance equations. A novel agent-based control strategy derived from economic models of markets is compared to two standard control strategies, proportional feedback and optimal control. The control systems require different amounts of knowledge about the underlying system dynamics, utilize different amounts of information about the current state of the system, and differ in their ability to achieve system-wide performance goals. Simulations illustrate the dynamic behavior of the life support system after it is perturbed away from its equilibrium state or nominal operating point under the three different control strategies. The performance of these strategies is discussed in the context of system-wide performance goals such as efficiency and robustness.


"Bilateral Trading on a Network: A Simulation Study"

This paper examines the dynamic behavior of a process of bilateral trading between spatially diverse agents. In contrast to the standard general equilibrium trading mechanism and to many previous models of bilateral trading, trades may only occur between agents who are located adjacent to one another. A network of bilateral trades provides one process by which an economy wide equilibrium can be achieved in a truly decentralized fashion.Computer simulations are used to explore the dynamic behavior of trading on a network. In particular, the speed of convergence to an economy equilibrium is examined for a variety of spatial structures. The process of convergence generates interesting intertemporal and cross-sectional dynamics such as the persistence of spatial correlations, or neighborhood effects, in the prices of the goods over time. In addition, the system exhibits path dependence: the equilibrium allocation and speed of convergence can depend crucially on details of the trading process such as the order of trades or location of agents on the network.