Edmund Durfee | University of Michigan (original) (raw)

Papers by Edmund Durfee

Most research on probabilistic commitments focuses on commitments to achieve conditions for other... more Most research on probabilistic commitments focuses on commitments to achieve conditions for other agents. Our work reveals that probabilistic commitments to instead maintain conditions for others are surprisingly different from their achievement counterparts, despite strong semantic similarities. We focus on the question of how the commitment recipient should model the provider’s effect on the recipient’s local environment, with only imperfect information being provided in the commitment specification. Our theoretic analyses show that we can more tightly bound the inefficiency of this imperfect modeling for achievement commitments than for maintenance commitments. We empirically demonstrate that probabilistic maintenance commitments are qualitatively more challenging for the recipient to model, and addressing the challenges can require the provider to adhere to a more detailed profile and sacrifice flexibility.

Proceedings of the AAAI Conference on Artificial Intelligence, 2020

Most research on probabilistic commitments focuses on commitments to achieve enabling preconditio... more Most research on probabilistic commitments focuses on commitments to achieve enabling preconditions for other agents. Our work reveals that probabilistic commitments to instead maintain preconditions for others are surprisingly harder to use well than their achievement counterparts, despite strong semantic similarities. We isolate the key difference as being not in how the commitment provider is constrained, but rather in how the commitment recipient can locally use the commitment specification to approximately model the provider's effects on the preconditions of interest. Our theoretic analyses show that we can more tightly bound the potential suboptimality due to approximate modeling for achievement than for maintenance commitments. We empirically evaluate alternative approximate modeling strategies, confirming that probabilistic maintenance commitments are qualitatively more challenging for the recipient to model well, and indicating the need for more detailed specifications ...

International Joint Conference on Artificial Intelligence, 2019

Proceedings of the Human Factors and Ergonomics Society ... Annual Meeting, Oct 1, 1997

Reducing the burden of interacting with complex systems has been a long standing goal of user int... more Reducing the burden of interacting with complex systems has been a long standing goal of user interface design. In our approach to this problem, we have been developing user interfaces that allow users to interact with complex systems in a natural way and in high-level, task-related terms. These capabilities help users concentrate on making important decisions without the distractions of manipulating systems and user interfaces. To attain such a goal, our approach uses a unique combination of multi-modal interaction and interaction planning. In this paper, we motivate the basis for our approach, we describe the user interface technologies we have developed, and briefly discuss the relevant research and development issues.

... References are not available. top of page CITED BY. Jeffrey Cox , Edmund Durfee, Efficient an... more ... References are not available. top of page CITED BY. Jeffrey Cox , Edmund Durfee, Efficient and distributable methods for solving the multiagent plan coordination problem, Multiagent and Grid Systems, v.5 n.4, p.373-408, December 2009. top of page INDEX TERMS. ...

support, patience, academic guidance, and technical advice, this dissertation would not have been... more support, patience, academic guidance, and technical advice, this dissertation would not have been possible. I especially want to thank Ed for giving me the freedom to explore and supporting me when I got excited about problems, even at times when he didn't quite share my enthusiasm. I truly believe Ed's top work priority is the growth of his students, and for that I offer my deepest thanks! I'm very grateful to the members of my committee, Kang Shin, Satinder Singh, Demos Teneketsis, and Michael Wellman for their valuable advice and insightful comments on my work. Michael Wellman deserves special credit for doing a very thorough job reviewing my thesis.

Journal of Artificial Intelligence Research, Apr 28, 2007

The judicious use of abstraction can help planning agents to identify key interactions between ac... more The judicious use of abstraction can help planning agents to identify key interactions between actions, and resolve them, without getting bogged down in details. However, ignoring the wrong details can lead agents into building plans that do not work, or into costly backtracking and replanning once overlooked interdependencies come to light. We claim that associating systematicallygenerated summary information with plans' abstract operators can ensure plan correctness, even for asynchronously-executed plans that must be coordinated across multiple agents, while still achieving valuable efficiency gains. In this paper, we formally characterize hierarchical plans whose actions have temporal extent, and describe a principled method for deriving summarized state and metric resource information for such actions. We provide sound and complete algorithms, along with heuristics, to exploit summary information during hierarchical refinement planning and plan coordination. Our analyses and experiments show that, under clearcut and reasonable conditions, using summary information can speed planning as much as doubly exponentially even for plans involving interacting subproblems.

Springer eBooks, 1988

A distributed problem solving network is composed of semi-autonomous problem-solving nodes that c... more A distributed problem solving network is composed of semi-autonomous problem-solving nodes that can communicate with each other. Nodes work together to solve a single problem by individually solving interacting subproblems and integrating their subproblem solutions into an overall solution. Because each node may have a limited local view of the overall problem, nodes must share subproblem solutions; cooperation thus requires intelligent local control decisions so that each node performs tasks which generate useful subproblem solutions. The use of a global “controller” to make these decisions for the nodes is not an option because it would be a severe communication and computational bottleneck and would make the network susceptible to complete collapse if it fails. Because nodes must make these decisions based only on their local information, well-coordinated or coherent cooperation is difficult to achieve [Davis and Smith, 1983; Lesser and Corkill, 1981].

Springer eBooks, 2013

Research into algorithms for coordinating computational agents that cooperatively solve problems ... more Research into algorithms for coordinating computational agents that cooperatively solve problems can shine light on potential strategies for coordinating human computation. Here, we briefly summarize key concepts manifested in distributed intelligent agent algorithms, and highlight some opportunities for translating pertinent concepts to benefit human computation.

Proceedings of the ... AAAI Conference on Artificial Intelligence, Apr 3, 2020

Cooperating agents can make commitments for better coordination, and commitments can only be prob... more Cooperating agents can make commitments for better coordination, and commitments can only be probabilistic when agents' actions have uncertain outcomes in general. Our perspective is that a commitment should be made not to outcomes but to courses of action. An agent thus earns trust by acting in good faith with respect to its committed courses of action. With this perspective, we examine an atypical form of probabilistic commitments called maintenance commitments, where an agent commits to actions that avoid an outcome that is undesirable to another agent. Compared with the existing probabilistic commitment framework for enablement commitments, our maintenance commitment poses new semantic and algorithmic challenges. We here formulate maintenance commitments in a decision-theoretic setting, examine possible semantics for how agents should treat such commitments, and describe corresponding planning methods. We conclude by arguing why we believe our efforts demonstrate that maintenance commitments are fundamentally different from enablement commitments, and what that means for their trustworthy pursuit.

Autonomous Agents and Multi-Agent Systems, Jan 18, 2020

The Kluwer international series in engineering and computer science, 1988

Having described how the planner works in the previous chapter, we now examine how the planner af... more Having described how the planner works in the previous chapter, we now examine how the planner affects problem solving. The first part of this chapter explores the activities of the planner and problem solver in a variety of experiments to better understand what the planner does. To fully understand the effects of the planner, these experiments not only examine how the planner improves control decisions, but also what the costs of those improvements are. It is important to remember that the planner’s job is to reduce the time needed to solve problems by improving control decisions, but if the planner needs a lot of time to make these decisions, then the net result may be that the time needs increase—the time saved in problem solving is used up in planning! In many of these experiments, therefore, the discussion covers not only how the planner affects local decisions but also whether the costs of planning are acceptable.

National Conference on Artificial Intelligence, 2015

Cooperating agents can make commitments to help each other, but commitments might have to be prob... more Cooperating agents can make commitments to help each other, but commitments might have to be probabilistic when actions have stochastic outcomes. We consider the additional complication in cases where an agent might prefer to change its policy as it learns more about its reward function from experience. How should such an agent be allowed to change its policy while still faithfully pursuing its commitment in a principled decision-theoretic manner? We address this question by defining a class of Dec-POMDPs with Bayesian reward uncertainty, and by developing a novel Commitment Constrained Iterative Mean Reward algorithm that implements the semantics of faithful commitment pursuit while still permitting the agent's response to the evolving understanding of its rewards. We bound the performance of our algorithm theoretically, and evaluate empirically how it effectively balances solution quality and computation cost.

Downsizing the number of operators controlling complex systems can increase the decision-making d... more Downsizing the number of operators controlling complex systems can increase the decision-making demands on remaining operators, particularly in crisis situations. An answer to this problem is to offload decision-making tasks from people to computational processes, and to use these processes to focus and expedite human decision making. In this paper, we describe a system comprised of multiple computational agents that has demonstrated an ability to help operators prioritize their tasks better, process their tasks faster, and enlist the aid of other operators more transparently. In developing this system, we have of course encountered challenges, particularly in devising content languages that adequately convey the right information (to be interpreted correctly) across the heterogeneous agents. We here summarize our work that addresses this challenge, and illustrate how our system improves performance for operators in naval situations.