Smith, D. E., 1985. Controlling inference, Stanford: Department of Computer Science, Stanford University, Ph.D. thesis.  Home/Search   Document Not in Database   Summary   Related Articles   Check

....other sorts of reasoning, one cannot assume that this metalevel decision can be made as easily. For example, in logical inference, it can be very difficult to estimate how frequently different inference goals occur, how frequently different methods succeed, and how much time these methods take. Smith (1986; 1988) has analyzed a number of important special cases of this problem, but much work remains to be done. But even when one does know how long inference operations take, allocating effort can be difficult, as is shown by Etzioni s (1991) work on rational control of traditional goal directed ....

Smith, D. E. 1986. Controlling inference. Technical Report STAN-CS-86-1107, Department of Computer Science, Stanford University.

....facts in the search space it generates. The problem is to focus the reasoner on the parts of the knowledge base that are relevant to the problem solving goal. Two major approaches in AI to overcoming this problem are meta level reasoning (e.g. Hayes 73] Genesereth 83] Genesereth 88] Smith 86] and creating abstraction hierarchies (e.g. Sacerdoti 74] Simon 81] and [Borgida and Etherington 89] The notion of irrelevance plays an important role in both fields in that a significant form of metalevel control knowledge are statements about irrelevance of elements of the ....

David Smith; "Controlling Inference"; Ph.D. thesis, Computer Science Department, Stanford university; 1986.

....(i.e. BC3) we only need to map every node in our search to a node in the query tree, and this can be done in a wide variety of inference algorithms. An interesting direction for future research is to combine the query tree 12 with other optimization methods. For example, Greiner [3] and Smith [13] addressed the issue of finding optimal strategies for rule and goal ordering. In doing so, they used a tree like representation of the possible derivations. The query tree can be used to extend their results to KBs with recursion and to incorporate the semantics of interpreted predicates. ....

David Smith. Controlling Inference. PhD thesis, Stanford University, Stanford, CA, 1986.

.... That is, the agent devotes a fraction of its effort to some simplified decisions about how to rationally expend the remainder, making this allocation by means of another rational choice involving restricted sorts of preferences, such as those expressed in default rules, or those employed by Smith (1985) and Russell and Wefald (1988) If need be, this simplification may be employed more than once to reach a decision. See (Doyle 1980, 1988a) for treatments of patterns of successively reflective preferences about preferences and reasoning about reasoning as elements of a theory of rationally ....

Smith, D. E., 1985. Controlling inference, Stanford: Department of Computer Science, Stanford University, Ph.D. thesis.

....a subset of the space that would have been searched by ordinary backward chaining, and therefore following the query tree enables pruning of parts of the search space. A different approach that was considered to speeding up inference is finding optimal strategies for searching a given space [Smith, 1986; Greiner, 1991; Greiner, 1992] The query tree can be used to complement and extend these methods in two ways. First, by delimiting the actual space that needs to be searched, some search paths can be eliminated from consideration when looking for the optimal search strategy. Second, the methods ....

.... ] 75 Partial evaluation of logic programs [ Smith and Hickey, 1990; Lloyd and Shepherdson, 1991; Bruynooghe et al. 1991 ] 75, 100 Tree automata [ Vardi, 1989 ] 74 Automated reasoning and query evaluation: Knowledge compilation [ Selman and Kautz, 1991 ] 100 Deriving optimal search strategies [ Smith, 1986; Greiner, 1991 ] 97 Message passing based query evaluation [ Van Gelder, 1986 ] 96 Magic set transformation [ Ullman, 1989; Mumick et al. 1990 ] 94 Independence of queries from updates: Detecting independence [ Blakeley et al. 1989; Elkan, 1990 ] 112, 126 Conjunctive query containment [ Klug, ....

Smith, David 1986. Controlling Inference. Ph.D. Dissertation, Stanford University, Stanford, CA.

....version of the meta IPS Programs similar to this IPS have already been developed. Elaine Kant s (1979) LIBRA system generates alternative algorithm refinements within the PSI automatic programming system, analyzes and compares their efficiencies, and selects the best refinement. David Smith s (1985) greedy algorithm for ordering of conjunctive queries performs a similar analysis. At each step, it considers which conjunct of the query should be processed (i.e. incorporated) next. It chooses the conjunct with the fewest estimated number of solutions. Since (in Smith s architecture) each ....

....most efficient base level problem solver, even if that search ends up taking more time than it would have taken to simply execute the unimproved baselevel program. This problem is common in all meta level architectures. The costs of doing the meta level reasoning may overwhelm the benefits (see Smith, 1985; Rosenschein Singh, 1983) The meta IPS solves this problem by restraining the IPS. Because the meta IPS seeks to minimize the combined cost of the IPS and the base level problem solver, it will always ensure that the cost of running the IPS is offset by the improvements to the base level ....

Smith, David E. 1985. Controlling Inference, Doctoral Dissertation, Department of Computer Science, Stanford University.

....has already been made by David Smith: Good control decisions are not arbitrary; there is always a reason why they work. Once these reasons are uncovered and recorded, specific control decisions will follow logically from the domainindependent rationale, and simple facts about the domain. [Smith,1985, p. 12] This is an observation with far reaching consequences, including the following: 1. There is no need for a metametalevel or anything along those lines. Domain independent control information need not be refined using higher order information. A similar conclusion has been reached in a ....

David E. Smith. Controlling Inference. PhD thesis, Stanford University, August 1985.

....of strong irrelevance, but it is a variation on weak irrelevance and is not an instance of computational irrelevance. The query tree encodes the space of possible derivations of the query. Recently, the question of finding optimal methods to search that space has received a lot of attention [ Smith, 1986, Greiner, 1991 ] a related issue is analyzing the utility of techniques in explanation based learning [ Etzioni, 1990, Greiner and Jurisica, 1992 ] Much of this work requires a graph like representation of the search space under consideration. The query tree is such a representation that ....

Smith, David 1986. Controlling Inference. Ph.D. Dissertation, Stanford Univesity, Stanford, CA.

.... of sorting [Horvitz, 1988] and of game playing search [Russell and Wefald, 1988, Hansson and Mayer, 1989] In related research on the control of logical inference, Smith, and Treitel and Genesereth, have explored the use of decision theory for selecting alternative logical reasoning strategies [Smith, 1986, Treitel and Genesereth, 1986] 3 Complexity of Inference In reasoning about real world actions under uncertainty, an agent generally must consider alternative decisions and outcomes, preferences about the possible outcomes, and the uncertain relationships among actions and outcomes. We have ....

Smith, D. (1986). Controlling inference. Technical Report STAN-CS-86-1107, Computer Science Department, Stanford University.

....be composed of many modules and the sequence and overlap of their execution is the major factor to achieve good performance. The concepts for arranging modules for optimal execution do not differ from those used in programming languages, databases [Dadaism] and artificial intelligence processing [Smith:86] However, such optimizers depend on models of the computations being performed. The models are evaluated with parameters being extracted or estimated, as the count of loop executions, the selectivity of a database predicate, the partitioning of the search space. Optimizing the execution of ....

David E. Smith "Controlling Inference" PhD Thesis, Stanford, CSD, CS-861107, Apr.1986.

No context found.

Smith, D. E., 1985. Controlling inference, Stanford: Department of Computer Science, Stanford University, Ph.D. thesis.

No context found.

David Smith. Controlling Inference. PhD thesis, Stanford University, Stanford, CA, 1986.

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