D. Einav and M. R. Fehling. Computationally-optimal real-resource strategies for independent, uninterruptible methods. In Uncertainty in Artificial Intelligence 6, pages 145--158. North-Holland, Amsterdam, 1991.  Home/Search   Document Not in Database   Summary   Related Articles   Check

....over the pre viously used myopic approaches to control computations. The latter select the next single computational step with the highest value as long as its comprehensive value is positive. One of the early examples of using non myopic control policies has been reported by Einav and Fehling [7]. They have developed an algorithm to construct generate and test policies to solve a problem with several given non interruptible methods. Unlike contract algorithms, these methods do not offer a tradeoff between computation time and quality. Another early example is the technique developed by ....

David Einav and Michael R. Fehling. Computationally-optimal real-resource strategies. IEEE International Conference on Systems, Man and Cybernetics, 581-586, 1990.

....of the value of computation. We formalize the meta level control problem as a sequential decision problem that can be solved by dynamic programming, in order to construct a non myopic solution. Dynamic programming has been used before for meta level control of computation. Einav and Fehling [7] and Russell and Subramanian [27] use dynamic programming to schedule a sequence of solution methods for a real time decision making problem; and Zilberstein, Charpillet and Chassaing [40] use dynamic programming to schedule a sequence of of contract algorithms to create the best interruptible ....

D. Einav, M.R. Fehling, Computationally-optimal real-resource strategies, in: Proc. IEEE International Conference on Systems, Man and Cybernetics, 1990, pp. 581--586.

.... is a limit only on the total time spent deliberating, but does not work well for a more complicated set of deadlines (e.g. the robot and the conveyor belt) 5 One of the problems with allocating deliberation time in discrete increments is that it is easy to wind up with a combinatorial problem [Einav, 1990]. This was part of our motivation in adopting the use of anytime decision procedures. If continuous behavior is a reasonable approximation, then deliberation scheduling is no longer a combinatorial problem, though it may still be computationally expensive for other reasons. Another problem is that ....

David Einav. Computationally-optimal real-resource strategies for inde- pendent, uninterruptible methods. In Proceedings of the Sixth Workshop on Uncer- tainty in Artificial Intelligence, pages 73-81, 1990.

....chunks are called methods. In [36] the computation to be controlled is modeled as a sequence of inference steps leading to the performance of an action. One of the problems with allocating deliberation time in discrete increments is that it is easy to wind up with a combinatorial problem [12]. An alternative approach is to assume that deliberation is performed using anytime decision procedures [10] also called flexible computations in [25] and that time can be allocated in values drawn from a continuous range (this is inevitably an approximation) Briefly, the advantages of this ....

Einav, David, Computationally-Optimal Real-Resource Strategies for Independent, Uninterruptible Methods, Proceedings of the Sixth Workshop on Uncertainty in Artificial Intelligence, Cambridge, MA, 1990, 73--81.

....can be quite powerful when used appropriately. This low execution cost was the motivation behind the randomized construction of configurationspace graphs in [97] The view presented here is inspired by the case based approach taken in [143] and is related to the decision making approaches in [7] [55]. Let M denote a space of basic motion planning problems, which can be generated by considering combinations of possible free configuration spaces (environments) initial configurations, and goal configurations. In practice, this set need not be enumerated. A state space, X, is defined as M ....

D. Einav and M. R. Fehling. Computationally-optimal real-resource strategies for independent, uninterruptible methods. In Uncertainty in Artificial Intelligence 6, pages 145--158. North-Holland, Amsterdam, 1991.

No context found.

D. Einav and M. R. Fehling. Computationally-optimal real-resource strategies for independent, uninterruptible methods. In Uncertainty in Artificial Intelligence 6, pages 145--158. North-Holland, Amsterdam, 1991.

No context found.

D. Einav and M. R. Fehling. Computationally-optimal real-resource strategies for independent, uninterruptible methods. In Uncertainty in Artificial Intelligence 6, pages 145--158. North-Holland, Amsterdam, 1991.

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