Eric J. Horvitz. Reasoning under varying and uncertain resource constraints. In Proceedings of the Seventh National Conference on Artificial Intelligence, pages 111--116, 1988.  Home/Search   Document Details and Download   Summary   Related Articles   Check

.... although in practice (c) improvements come incrementally (after [12] Given expected utilities for each partial result, algorithms known as deliberation schedulers have been developed which optimally schedule the use of computational resources to maximize the expected utility of the result [13, 12]. The scheduler used in [10] greedily rendered those scene components with the highest utility gain associated with them, that could be completed in the allotted time. When applied to cue integration, a deliberation scheduler must choose between the cues, to decide which will receive further ....

....this idea in realtime implementations requires careful management of resources to get the most out of using each cue: meta reasoning. The next step is to develop a deliberation scheduler suitable for cue integration. To optimize the scheduling, meta reasoning approaches use expected utilities [13] from each computation being considered to make the best use of resources. In other words, the scheduler attempts to predict which computation would provide the greatest benefit. In most meta reasoning research [13, 12] utilities of inputs are simply combined additively to determine the utility ....

[Article contains additional citation context not shown here]

Horvitz, E.: Reasoning under varying and uncertain resource constraints. In: AAAI '88. (1988) 111--116

....to solving the problem of searching in the plane. Furthermore, the analytical work reported here offers a dramatic simplification of the proofs presented in [Baeza Yates et al. 1993] Sections 4 and 5 build on a large body of work on meta level control of computation by Dean and Boddy [1988] Horvitz [1988], Russell and Wefald [1991] and others. By and large, existing meta level control mechanisms are myopic; the computation is terminated once no single computational step has positive value. One exception is the technique developed by Russell, Sub ramanJan, and Parr [1993] for sequencing a set of ....

Eric Horvitz. Reasoning under varying and uncertain resource constraints. Seventh National Conference on Artificial Intelligence, 111-116, 1988.

....resources has been studied extensively by the AI community since the mid 1980 s. These efforts have led to the development of a variety of techniques such as anytime algo rithms [Dean and Boddy, 1988; Zilberstein and Russell, 1996] design to time [Gravey and Lesser, 1993] flexible computation [Horvitz, 1988], imprecise computation [Liu et al. 1991] and progressive reasoning [Mouaddib, 1993; Mouaddib and Zilberstein, 1997] We adopt the progressive processing framework to formalize and solve the meta level control problem of a realtime information retrieval application. The technique maps each task ....

Eric Horvitz. Reasoning under varying and uncertain resource constraints. Seventh National Conference on Artificial Intelligence, 111-116, 1988.

....research grant, and by the LIRE Cooperative Program at INKIA. 1. Introduction Since the mid 1980 s, the artificial intelligence (AI) research community has produced a large body of work on incremental problem solving techniques such as anytime algorithms [4,18] and flexible computation [10,11]. Numerous such algorithms have been constructed for solving core AI problems such as heuris tic search, constraint satisfaction, planning and scheduling, and diagnosis. The working notes of the 1996 AAAI Fall Symposium on Flexible Computation offer a good sample of such techniques and ....

....computational resources and react to dynamic changes in their environment. When combined with an appropriate meta level control, anytime algorithms make it possible to build systems that optimize the amount of deliberation based on the actual progress they make and the urgency to take action [4,11]. Some anytime algorithms, however, are not interruptible. Such algorithms, called contract algorithms [18] require the amount of run time to be determined prior to their activation. In other words, contract algorithms offer a tradeoff between computation time and quality of results, but they do ....

[Article contains additional citation context not shown here]

Eric Horvitz. Reasoning under varying and uncertain resource constraints. Seventh National Conference on Artificial Intelligence, 111-116, 1988.

....general model in which utility also depends on the state of a dynamic environment. Definition 2. A time dependent utility q,t) represents the utility of a solution of quality q at time t . It is often possible to simplify this function by treating it as the sum of two functions that Horvitz [14] calls object level utility and inference related utility. Object level utility represents the utility of a solution without regard to the costs associated with its computation, and inference related utility represents these computational costs. We adopt the following terminology of Russell and ....

E.J. Horvitz, Reasoning under varying and uncertain resource constraints, in: Proc. AAAI-88, St. Paul, MN, 1988, pp. 111--116.

....the probability of evidence is very low [Ch01] Many model simplification methods and search based approaches are also anytime algorithms. In late 1980s Eric Horvitz first investigated the problem of uncertain reasoning under limited computational resources under the name of flexible computation [Ho87, Ho88, Ho90]. His bounded conditioning algorithm was the first anytime Bayesian network inference algorithm (under the name of flexible computation) HSC89] Bounded conditioning uses conditioning method, but conditions only on a small, high probability cutset instances. As more resources (time) are ....

E. Horvitz. Reasoning under varying and uncertain resource constraints. Proceedings of the Seventh National Conference on Artificial Intelligence, Minneapolis, MN. August 1988. Morgan Kaufmann, San Mateo, CA. pp. 111-116, 1988.

....by looking it up in the projection. As was demonstrated in the example, we developed other mechanisms for a planner that reason in time which we don t describe here. Our approach has many concerns in common with existing research in planning and temporal reasoning. McDermott, 1982, Haas, 1985, Horvitz, 1988, Horvitz et al. 1989, Russell and Wefald, 1989 ] However, these works do not account for the time taken for meta planning. Indeed, this is stated in [ Russell and Wefald, 1989 ] page 402) Here we will not worry about the cost of meta reasoning itself; in practice, we have been able to ....

E. J. Horvitz. Reasoning under varying and uncertain resource constraints. In Proceedings of the 7th National Conference on Artificial Intelligence, pages 111--116, St. Paul, MN, 1988. AAAI.

.... As a result most of those methods resort to simplifying assumptions such as myopic deliberation control [3,28,29] conditioning the deliberation control on hand picked features [28,29] assuming that an algorithm s future performance can be deterministically predicted using a performance profile [10,11], assuming that an anytime algorithm s future performance does not depend on the run on that instance so far [4,9,38, 39] or that performance is conditioned on quality so far but not the path [8] or resorting to asymptotic notions of bounded optimality [27] While such simplifications can be ....

E.J. Horvitz, Reasoning under varying and uncertain resource constraints, in: Proc. AAAI-88, St. Paul, MN, Morgan Kaufmann, San Mateo, CA, 1988, pp. 111--116.

.... Applications We have already mentioned several application areas in which the decision theoretic control of reasoning has been explored, including planning [Elkan, 1990, Drummond and Bresina, 1990] game tree search [lussell and Wefald, 1989a, Hansson and Mayer, 1988] medical decision making [Horvitz, 1988], and robot control [Dean and Boddy, 1988, Boddy and Dean, 1989] In addition, decision theoretic deliberation scheduling frameworks have been proposed for computer vision [Levitt et al. 1988] industrial control [Agogino et al. 1988, lamamurthi and Agogino, 1988] probabilistic inference using ....

....1990] sensor fusion [Hager, 1988] and queries in deductive databases [Smith, 1989] Breese and Febling [1988] provide an abstract architecture for the decision theoretic control of an autonomous agent, including reasoning about tradeoffs among planning, acting, and gathering information. [Horvitz el a. 1988] provide a survey of the ap plication of decision theory to AI more generally, and points out some areas in which further research might be most fruitful. 1.4 Related Work in Other Areas The kinds of optimization problems we address are similar to problems addressed in a number of other ....

[Article contains additional citation context not shown here]

Eric J. Horvitz. Reasoning under varying and uncertain resource con- straints. In Proceedings AAAI-88, pages 111-116. AAAI, 1988.

....above tacitly assume that a task s output is of no value if it is not executed completely. However, in many application areas such as multimedia applications [17] image and speech processing [4, 6, 19] timedependent planning [3] robot control navigation systems [21] medical decision making [9], information gathering [7] realtime heuristic search [12] and database query processing [20] a partial or approximate but timely result is usually acceptable. The imprecise computation [5, 15] and IRIS (Increased Reward with Increased Service) 10, 13] models were proposed to enhance the ....

E.J. Horvitz. Reasoning under varying and uncertain resource constraints Proceedings of the Seventh National Conference on Artificial Intelligence, AAAI-88, pp. 111-116, August 1988.

....goals and plans using mathematical logic, and attempted to define a prescriptive theory of action using the same tool. An alternative approach, placing more emphasis on decision theoretic considerations (so as to ensure reasoned choice among alternative strategies) is represented, e.g. in [FS] [Hz], Do] DB] De] and the book [RW] Computers playing classical games are already a field in itself with special purpose computers and specialized periodicals. Two nontraditional games that may be worth mentioning are Diplomacy, for which a program was developed [KEL] which negotiates with the ....

E. J. Horvitz, Reasoning Under Varying and Uncertain Resource Constraints, in Proceedings of the Seventh National Conference on Artificial Intelligence (1988) 111 - 116. 51

....of a previously unknown SAM site on its exit path. The agent will generate a new controller for the egress phase of its mission that will be able to handle this threat. Due to bounded resources, the opportunity of dynamic adaptation poses the corresponding problem of deliberation scheduling [1, 4]. The SA CIRCA agent must determine how best to allocate its limited computational resources to improving controllers for various mission phases. For example, should the agent first improve the controller for the final phase, since it is perceived to be 2 ; to evade simple radar missiles, start ....

....and the average loss of the simple greedy agent was 15.2 . 6 Related Work 6.1 Anytime Algorithms To the best of our knowledge, the term deliberation scheduling was introduced by Boddy and Dean. There was a great deal of work in this area around the early 1990s by Boddy and Dean [1] 13 Horvitz [4, 3], and Russell and Wefald [10] These researchers (and others) investigated methods for using decision theory to address the problem of managing computation under bounded resources. Boddy and Dean [1] categorize deliberation scheduling as being of two sorts: either discrete or anytime. In the ....

E. J. Horvitz, "Reasoning under varying and uncertain resource constraints," in Proceedings of the Seventh National Conference on Artificial Intelligence, pp. 111--116, Los Altos, CA, 1988, Morgan Kaufmann Publishers, Inc.

....tacitly assume that a task s output is of no value if it is not executed completely. However, in many application areas such as multimedia applications [26] image and speech processing [5, 6, 9, 28] time dependent planning [4] robot control navigation systems [12, 30] medical decision making [13], information gathering [10] real time heuristic search [17] and database query processing [29] a partial or approximate but timely result is usually acceptable. The imprecise computation [7, 19, 21] and IRIS (Increased Reward with Increased Service) 14, 15, 18] models were proposed to enhance ....

E.J. Horvitz. Reasoning under varying and uncertain resource constraints Proceedings of the Seventh National Conference on Artificial Intelligence, AAAI-88, pp. 111-116, August 1988.

....description. The predictive model is then used to help guide the solver on future problem instances by making choices that minimize the expected time to solution. The general framework is illustrated in Fig. 4. This effort builds on work on resource bounded reasoning and the value of computation [40, 39, 80], as well as the large body of work on learning Bayesian models [21, 14] Our initial results [41] are quite promising, and show a significant degree of power in predicting the run time of CSP and SAT engines for solving QCP problems after examining traces of the first few hundred branches in the ....

E. Horvitz. Reasoning under varying and uncertain resource constraints. In Proceedings of AAAI-88, pages 111--116. Morgan Kaufmann, 1988.

No context found.

E.J. Horvitz. Reasoning under varying and uncertain resource constraints. In Proceedings AAAI-88 Seventh National Conference on Artificial Intelligence, pages 111--116. American Association for Artificial Intelligence, August 1988.

....monitored the progress of search in a propositional theorem prover and used measures of progress in updating the probability of truth or falsity of assertions. Stepping back to view the larger body of work on the decision theoretic control of computation, measures of expected value of computation [15, 8, 25], employed to guide problem solving, rely on forecasts of the re nements of partial results with future computation. More generally, representations of problemsolving progress have been central in research on exible or anytime methods procedures that exhibit a relatively smooth surface of ....

E.J. Horvitz. Reasoning under varying and uncertain resource constraints. In Proceedings of AAAI-88, pages 111-116. Morgan Kaufmann, San Mateo, CA, August 1988.

....but can expect to eventually face events and challenges that may lead to real time computational bottlenecks. Research on continual computation comes in the spirit of a body of work, centering on the value of optimizing the performance of reasoning and decision making under limited resources [7,11,22,26,27,29,43,46,48,51,59,63,68]. The work also shares motivations and goals with a variety of related efforts on compilation, precomputation, and prefetching in Computer Science. Policies for guiding the precomputation and caching of complete or partial solutions of potential future problems are targeted at enhancing the ....

....computation strategies has explored the application of decision theory to reason about the expected value of allocating resources to refine partial results in different settings. These efforts typically assume a utility model for assigning value to the results and costs for allocated resources [6,48]. A key construct used in studies of deliberation of flexible procedures is the expected value of computation (EVC) 47,48,51,61] Definition 4. The expected value of computation (EVC) is the change in the net expected value of a system s behavior with the refinement of one or more results by ....

[Article contains additional citation context not shown here]

E.J. Horvitz, Reasoning under varying and uncertain resource constraints, in: Proc. AAAI-88 St. Paul, MN, Morgan Kaufmann, San Mateo, CA, 1988, pp. 111--116.

....A Bayesian model was harnessed to update belief about di erent outcomes as a function of the amount of time that problem solving continued without halting. Stepping back to view the larger body of work on the decision theoretic control of computation, measures of expected value of computation [15, 8, 25], employed to guide problem solving, rely on forecasts of the re nements of partial results with future computation. More generally, representations of problemsolving progress have been central in research on exible or anytime methods procedures that exhibit a relatively smooth surface of ....

E.J. Horvitz. Reasoning under varying and uncertain resource constraints. In Proceedings of AAAI-88, pages 111-116. Morgan Kaufmann, San Mateo, CA, August 1988.

No context found.

Eric J. Horvitz. Reasoning under varying and uncertain resource constraints. In Proceedings of the Seventh National Conference on Artificial Intelligence, pages 111--116, 1988.

No context found.

E. J. Horvitz. Reasoning under varying and uncertain resource constraints. In Proc. National Conf. American Association for Artificial Intelligence (AAAI-88), pages 111--116, 1988.

No context found.

E. J. Horvitz. Reasoning under varying and uncertain resource constraints. In Proceedings of the National Conference on Artificial Intelligence (AAAI), pages 111--116, St. Paul, MN, August 1988. Morgan Kaufmann, San Mateo, CA.

No context found.

E. J. Horvitz. Reasoning under varying and uncertain resource constraints. In Proceedings of the Seventh National Conference on Artificial Intelligence, pages 111-- 116. AAAI Press, August 1988.

No context found.

Eric J. Horvitz. Reasoning under varying and uncertain resource constraints. In Proceedings of AAAI'88, pages 111--116, St. Paul, MN, August 1988.

No context found.

E. Horvitz, Reasoning under varying and uncertain resource constraints. Proceedings of the Seventh National Conference on Artificial Intelligence, Minneapolis, MN. August 1988. Morgan Kaufmann, San Mateo, CA. pp. 111-116, 1988.

No context found.

E. Horvitz. Reasoning under varying and uncertain resource constraints. In Uncertainty in Artificial Intelligence (UAI-88), pages 111--116, 1988.

First 50 documents  Next 50

Online articles have much greater impact   More about CiteSeer.IST   Add search form to your site   Submit documents   Feedback  

CiteSeer.IST - Copyright Penn State and NEC