N.J. Nilsson. Principles of Arti#cial Intelligence. Tioga Publ., Palo Alto, CA, 1980.  Home/Search   Document Not in Database   Summary   Related Articles   Check

....is built on a more fundamental inference step: nding the r highest scoring substitutions for a conjunctive query Q. Below we will call these r topscoring substitutions the r answer to Q. The algorithm for eciently nding an r answer is based on a combination of admissible A search [35, 25] with IR indexing and pruning methods [39, 42] Below, we will brie y describe the algorithm for answering a conjunctive query Q. More detail on WHIRL s inference algorithm, including examples and a proof of correctness, can be found elsewhere [7] In the query answering algorithm, high scoring ....

....value f(s) If s is a goal state (i.e. a ground substitution) then it is added to the set of answers for the query. Otherwise, one of the operators described above is use to nd the children of s, which are then added to the OPEN list. This algorithm is a variant of the usual formulation of A [35, 25]. For this search strategy to be admissible (i.e. guaranteed to output the best r answers in order of non decreasing score) the goodness function f(s) for s = hQ; Ei must be an upper bound on the score of any ground substitution reachable from that state. To obtain this upper bound we let ....

Nils Nilsson. Principles of Arti cial Intelligence. Morgan Kaufmann, 1987.

....ASPECTS OF RULE BASED REASONING Finally, after the xpoint theory and the proof theory of quantitative deduction, we consider its game theoretic aspects. We rst review the main concepts of twoperson games, because these have close parallels to rule based reasoning (see, for example, [6, 7]) both qualitative and quantitative. These parallels suggest an algorithm for a quantitative version of a prolog interpreter. There are two players, White and Black, and there is a state (for example the disposition of pieces on a board; or of matches over heaps, as in Nim) Starting from the ....

Nilsson, N. J., Principles of Arti cial Intelligence, Tioga, 1980.

....general purpose [30, 46] and application speci c [44, 55, 73] The underlying techniques are described in the next section. 3. 1 Keygraph Searching in Graph Databases Cook et al. 30, 35] applied an improvement of the inexact graph matching method (algorithm A ) described by Nilsson [79] based on an inexact graph matching algorithm proposed in [21] to nd similar repetitive subgraphs in a single graph database. Thus, their methods are primarily of interest for the third step above. Their system, SUBDUE, has been applied to discovery and search for subgraphs in protein databases, ....

....can be classi ed as approximate, inexact, and exact algorithms. Approximate algorithms [6, 27, 37, 43, 91, 101] have polynomial complexity but they are not guaranteed to nd a correct solution. Exact and inexact algorithms do nd correct answers and therefore have exponential worst case complexity [14, 48, 50, 63, 74, 79, 82, 102]. Inexact algorithms employ error correction techniques for a noisy data graph. These algorithms employ a cost function to measure the similarity of the graphs. For example, a cost function may be de ned based on semantic or syntactic transformations to transform one graph into another. Of ....

[Article contains additional citation context not shown here]

N. Nilsson. Principles of Articial Intelligence. Tioga, Palo Alto, California, 1980.

....algorithms and a formal model for according heuristic search that takes into account the speci cs of box decomposition methods, without restricting itself to unnecessary algorithmic details. This makes the result better suited for our problems than the classical models from Arti cial Intelligence [33, 30]. We show the relevance of the developed framework by showing how and why the box choice heuristics available in the literature for global optimization [29, 39, 7] can be applied to approximate quanti ed constraint solving [35] The author believes that the presented framework provides an ideal ....

.... ( predicate symbols (e.g. function symbols (e.g. sin, exp) rational constants and variables ranging over the reals) nd its solution set (or truth value in the case of a closed formula) A naive application of game tree search methods from Arti cial Intelligence [33, 30] to minimax optimization would start from the observation that a minimax optimization problem represents a game tree for which each tree level represents a certain variable, each node has in nitely many children (one for each substitution of a real constant for the according variable) and the ....

[Article contains additional citation context not shown here]

N. J. Nilsson. Principles of Articial Intelligence. Springer-Verlag, 1982.

....the Min CVCB problem and its variations (e.g. 7, 9, 16, 17] Most of these algorithms are heuristic and have no guaranteed performance. An algorithm proposed in [7] introduced the concept of critical set and used branch and bound technique based on the A algorithm in Arti cial Intelligence [14]. No explicit analysis was given in [7] for this algorithm, but it is not hard to see that in the worst case the algorithm running time is at least (2 ku k l m p n) Experimental results show that this algorithm is much more favourable than previous algorithms and the running time is ....

N. J. Nilsson, Principles of Articial Intelligence, Tioga Publishing Co., 1980.

....propositional natural deduction. FB is the logic of a theorem prover which supports interactive proof construction in general domains. 1 Introduction Forward and backward reasoning are well known techniques applied in both knowledge representation and automated theorem proving (see for instance [17, 4, 19, 15, 25, 14, 8]) Basically, any problem can be solved in a forward direction, i.e. from the hypothesis to the conclusion, or in a backward direction, i.e. from the goal to (sub)goals. Forward and backward reasoning can be integrated in a single bidirectional reasoning system, where a deduction can be performed ....

....logic. At the rst level, which we call the heuristic level, bidirectional reasoning is seen as an ecient strategy for problem solving, and it is discussed and formalized almost independently of the underlying logic. Some examples in mathematics are [22, 23] and [27] One example in AI is [19]. At this level, it does make sense to speak about the psychological plausibility of bidirectional reasoning (see for instance [15, 25, 3] At the second level, which we call the tactic level, bidirectional reasoning is performed by using rules (called tactics) which are built on the underlying ....

[Article contains additional citation context not shown here]

N.J. Nilsson. Principles of Articial Intelligence. Tioga Publishing Co., 1980.

....assumption, CWA) is given in table 2.1. Each operator has pre conditions (PRE) which must hold in a current state if the operator is applied. E ects are represented by ADD and DEL lists, specifying which literals are satis ed after operator application and which literals are no longer satis ed (Nilsson 1980). Putting a block x on another block y is splitted into two operators one for the case, that x is currently lying on the table (put 1 in gure 2.2) and one for the case, that x is lying on another block z (put 2 in gure 2.2) For a given goal fON(A, B) ON(B, C) ONTABLE(C)g the minimal ....

.... ( the source node of the arc labelled by the operation op i which makes P i true) is labelled by P i 1 ; P i 2 ; P i k ; P 1 ; P 2 ; P n ; with P i deleted in the sequence and its sub goals P i 1 ; P i k (if there are any) placed in front of the sequence (Nilsson 1980). The goal sequence has a stack like structure but is not a stack in the proper sense. 2. If P i is the result of a type 2 transition occuring in (some branch of) the subtree with the current node as root and is relevant at the current node then nothing is to be done. Being relevant means ....

Nilsson, N. J. (1980). Principles of Articial Intelligence. New York: Springer.

....(Sect. 5) We nally discuss implementation related issues and re nements, and related work (Sect. 6 and 7) 2 2 Branch and Bound in Combinatorial Optimization The notion of lower bounds used in Combinatorial Optimization [1, 12] is familiar in AI where they are called admissible heuristics [32, 34]. Branching, on the other hand, is a less familiar notion, and the word does not even appear in the index of AI textbooks. This is probably due to the close correspondence between branching and action application in the search problems most often considered in AI. Indeed in problems like the ....

....B 2 R(C) i for some action a s.t. add(a) C 6= and del(a) C = B = C add(a) pre(a) The notion of regression is familiar in AI and corresponds to the inverse application of actions; namely, B 2 R(C) i B is a minimal set of atoms that would lead to C by applying one of the actions [32]. 6 The higher the value of m, the more accurate the heuristic but the more expensive its computation. The computation of h m is similar to the computation of shortest paths in a graph whose nodes are the di erent atom sets of size equal to or smaller than m. Thus its complexity is a low ....

N. Nilsson. Principles of Articial Intelligence. Tioga, 1980.

....3 and 4 we give some notes on the MasPar in section 5. In section 6, we present some signi cant experiments and results. We conclude with a summary in section 7. 2 The A Algorithm for the Optimization of the Network Topology The A is a heuristic optimization method used in graph theory [Nil82]. Let G= nodes ; edges) be a graph de ned by an initial node x 0 and an expansion rule : 2 where nodes and edges Here, denotes the set of all possible fully connected, feedforward network topologies without shortcuts. Given an heuristic function h : R, a cost function g : R, and a ....

....node x 0 and an expansion rule : 2 where nodes and edges Here, denotes the set of all possible fully connected, feedforward network topologies without shortcuts. Given an heuristic function h : R, a cost function g : R, and a solution criterion S : Bool with certain properties [Nil82], the A nds a solution xL2fx2 : S(x)g. This solution is optimal in the sense that g(xL ) min (x2 S(x) g(x) To nd this solution, the A investigates the minimal number of nodes x 2 Figure 1 a) shows the algorithm using pseudocode. To use the A for network topology optimization, we de ne ....

N. Nilsson. Principles of Articial Intelligence. Springer-Verlag, 1982.

....X = OE 1 c 1; X = OE X = OE :7 Theta minf1g r p(X) X = OE maxf:7g X = OE 1 c 2; X = OE X = OE :5 Theta minf1g r p(X) X = OE maxf:5g Clearly, X = OE :7 p(X) X = OE is a logical consequence of PF . As proposed by [26] search strategies such as alpha beta pruning (see [22]) can be used quite directly to dene an interpreter for quantitative rule sets. The same techniques can be applied to a min max proof procedure in quantitative CLP. In general, the amount of search needed to nd the best proof for a goal, i.e. the maximal valued proof tree for a goal from a ....

Nils J. Nilsson. Principles of Articial Intelligence. Springer, Berlin, 1982.

....local maps are grouped together to dene the local map validity regions. Along with the perception uncertainty eld puf, a local map validity region R and its local map F dene a landmark . 6 Safe Path Planning The roadmap, which is a graph, is explored using the classical Dijkstra algorithm [15]. This algorithm is guaranteed to return the path that optimizes a given cost function. Path length is a straightforward optimality criterion. Like [18] it proved interesting though to dene a cost function combining path length and path reliability. The cost function is detailed in #6.1. In the ....

N. J. Nilsson. Principles of articial intelligence. Morgan Kaufmann, Los Altos, CA (USA), 1980.

....problems for benchmarking is a current phenomenon related to STRIPS descending planning systems. As one of the anonymous reviewers pointed out to us, quite a number of non invertible planning problems have also been proposed in the planning literature, e.g. the register assignment problem (Nilsson, 1980), the robot crossing a road problem (Sanborn Hendler, 1988) some instances of manufacturing problems (Regli, Gupta, Nau, 1995) and the Yale Shooting problem (McDermott Hanks, 1987) For these problems, i.e. for problems that are not invertible, one could in the spirit of argument 1 at the ....

Nilsson, N. (1980). Principles of Articial Intelligence. Tioga Publishing Company, Palo Alto.

....same local maps are grouped together to dene the local map validity regions. Along with the perception uncertainty eld puf, a local map validity region R and its local map F dene a landmark . 6 Safe Path Planning The roadmap, which is a graph, is explored using the classical Dijkstra algorithm [15]. This algorithm is guaranteed to return the path that optimizes a given cost function. Path length is a straightforward optimality criterion. Like [18] it proved interesting though to dene a cost function combining path length and path reliability. The cost function is detailed in #6.1. In the ....

N. J. Nilsson. Principles of articial intelligence. Morgan Kaufmann, Los Altos, CA (USA), 1980.

....introduced in [10] with motivations related to two level grammars. Later, the investigation of CD grammar systems became a vivid area of research after relating CD grammar systems with Arti cial Intelligence (AI) notions [1] such as multi agent systems or blackboard models for problem solving [12]. From this point of view, motivations for CD grammar systems can be summarized as follows: Several grammars (agents or experts in the framework of AI) mainly consisting of rule sets (corresponding to scripts the agents have to obey to) are cooperating in order to work on a sentential form ....

N. J. Nilsson. Principles of Articial Intelligence. Berlin: Springer, 1982.

....the new goal is used to adjust (lower) the upper bound and the search continues. Depth rst branch and bound depends on a high goal density, otherwise it will su er from the same problems as depth rst search. Bidirectional Search Nothing forces us to solve the problem in a forward direction [Nil80] Why not search backwards , starting from the goal state and attempting to nd a path to the start state Choosing the right direction ( forward or backward ) can lead to signi cant savings, since tree shapes might not be symmetric and a forward tree might be larger than the corresponding ....

N. Nilsson. Principles in Articial Intelligence. Morgan Kaufman Publisher, Inc., Tioga, Palo Alto, CA, 1980.

....domain. This Situation Driven Selection paradigm is described in section 4.1. From this paradigm, a speci cation language has been derived. This language is described in section 4.2. The execution engine that the CSM uses is described in section 4.3. This engine was inspired by the STRIPS system ([Nil80]) one of the earliest planning systems that could deal with compound goals. Section 4.4 explains how the CSM interfaces with the rest of the system. Chapter 5 will show how this system was used to control a robot and gives also the results of the completed experiments. 4.1 Situation Driven ....

Nils J. Nilsson. Principles of Articial Intelligence. Tioga, Palo Alto, 1980.

No context found.

N.J. Nilsson. Principles of Arti#cial Intelligence. Tioga Publ., Palo Alto, CA, 1980.

No context found.

N. Nilsson. Principles of Arti cial Intelligence. Tioga Publishing Company, 1980.

No context found.

N. J. Nilsson. Principles of Arti cial Intelligence. Springer-Verlag, 2nd edition, 1982.

No context found.

N J Nilsson. Principles of Articial Intelligence. Tioga, Palo Alto, 1980.

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