J. Laird. Full abstraction for functional languages with control. Extended abstract, in the Proceedings of the 12th Annual Symposium on Logic in Computer Science, LICS '97, 1997.  Home/Search   Document Details and Download   Summary   Related Articles   Check

.... an Overview semantics: the state of the art Games models have been built for higher order programming languages with a variety of computational features: pure functional languages (PCF) 1, 8, 12] store (Idealized Algol) 3] operators (SPCF, exceptions) [9] . store (pointers) 2] 7] Semantics: an Overview Full abstraction many cases, the models completely characterize those behaviours which can be programmed in the language: results hold. Whenever a definability result holds, it is possible to achieve full ....

Jim Laird. Full abstraction for functional languages with control. In Proceedings, Twelfth Annual IEEE Symposium on Logic in Computer Science, pages 58--67. IEEE Computer Society Press, 1997.

.... cant point in this context is the close connection between the presented calculus and game semantics [3, 22, 25] the structure of interaction of typed processes (with respect to typed environments) precisely conforms to the intensional structures of games introduced in [25] and studied in e.g. [2, 12, 22, 27, 28]. It is notable that the type discipline itself does not mention basic notions in game semantics such as visibility, well bracketing and innocence (although it does use a syntactic form of IO alternation) yet they are derivable as operational properties of typed processes. We use this ....

Laird, J., Full abstraction for functional languages with control, LICS'97, IEEE, 1997.

....for the interpretation of sums to follow. The latter is less fundamental for the results of the present paper, but in more recent work, Laird has fruitfully exploited the decoupling of the notion of view from the question answer structure to model languages with non local control operators [23]. A justified sequence s is well formed if it satisfies (w1) Players alternate: if s = s 1 mns 2 then (m) 6= n) w2) The bracketing condition . We say that a question q in s is answered by a later answer a in s if q justifies a. The bracketing condition is satisfied by s if for each ....

.... the innocence condition gives a fully abstract model of Idealised Algol, a prototypical imperative language with procedures and local variables [6] while relaxing the bracketing condition, but retaining innocence, gives a fully abstract model of PCF extended with the control operator call cc [23]. This may also lead to a better understanding of the connections between game semantics and the intensional semantics provided by sequential algorithms [8] ....

J. Laird. Full abstraction for functional languages with control. Draft paper, December 1996.

....of one of these constraints on strategies, leading to a larger category. Thus for example G b is the category of well bracketed but not necessarily innocent strategies. Remarkably, there is a precise correspondence between the syntactic and semantic squares, as shown in a series of papers [4,6,8,22]. For example, relaxing the constraint of innocence allows local state to be modelled, while relaxing bracketing allows control G i G G ib G b Figure 2: The semantic square operators to be modelled. Moreover, these increments in expressive power are exact, as shown by factorization theorems. ....

....We can now prove definability for the model of SPCF in the category C i . As was the case for IA, the proof makes use of a factorization theorem which removes violations of the bracketing condition from an arbitrary compact innocent strategy. The technique behind this factorization is due to Laird [22]. Lemma 39 Let A be an SPCF type and oe : A an innocently compact strategy. There exists a natural number k and an innocently compact, well bracketed strategy : N 1 Theta Delta Delta Delta Theta N k ) N k 1 ) N k 2 ) N such that c ; oe, where c k is the strategy used in the ....

J. Laird. Full abstraction for functional languages with control. In Proceedings, Twelfth Annual IEEE Symposium on Logic in Computer Science, pages 58--67. IEEE Computer Society Press, 1997.

.... urry of remarkable theoretical results showed that games based techniques can give fully abstract models for pure functional languages [2,14,16] imperative languages with local variables and higher order procedures [3,4] languages with general references [1] languages with generalized control [15] and more. From a practical perspective, the game models have two attractive features. First, they are fully formal. The moves can be represented by elementary, uninterpreted, symbols; the rules are combinatorial in nature. This means that game models need not deal with the deep semantic issues ....

Laird, J., Full abstraction for functional languages with control, in: Proceedings, Twelth Annual IEEE Symposium on Logic in Computer Science, IEEE Computer Society Press, Warsaw, Poland, 1997, pp. 58-67.

....category of non deterministic games in [6] The dotted arrows are recovered by extending this functor to operate on parallel composition. Within this underlying framework we can also model call by value languages [5] untyped languages [15] and languages with control features such as exceptions [14]. In the absence of higher order features, the basic notion of security is that, once program variables are given a security clearance, no user able to read values stored in variables with clearance c can gain any information about values stored in variables with a clearance higher than c. Many ....

....Nikau model [13, 21] We will denote by C cbn the subcategory of [13] with objects well opened games where the intensionally fully abstract model lives. More recent full abstraction results (for Idealised Algol[5] Idealised Algol with general references [2] PCF extended with control [14]) have been obtained by carefully relaxing some of the conditions of innocence or bracketing. For example knowing strategies (i.e. strategies that are not innocent) which are well bracketed are sufficient 1 Here A is the unbounded tensor game which has as moves the disjoint union of omega ....

[Article contains additional citation context not shown here]

J. Laird, Full abstraction for functional languages with control. In Proc. LICS'97, IEEE Press, 1997.

....as probabilistic strategies are extremely natural in VonNeumann matrix games. Indeed, our category smoothly extends the deterministic games world charted out by Hyland Ong [13] and Nickau [20] and further developed by Abramsky McCusker [3, 4, 5, 19] Abramsky, Honda McCusker [2] and Laird [15]. The category also fits in very nicely with the basic concepts of probability theory. The interaction of a strategy and a counter strategy on a game gives rise to a Borelian subprobability on the infinite plays. The category has a tensor product which corresponds to the classical product measure ....

J. Laird. Full abstraction for functional languages with control. In Proceedings, Twelfth Annual IEEE Symposium on Logic in Computer Science. IEEE Computer Society Press, 1997.

....Subject to some technical condition about Answers (that they can t be pointed at) we prove a suitable form of conditional named case is a universal oracle with respect to rigidity. A result parallelizing Laird s proof that a form of taster known as catch is an oracle for bracketedness in [Lai97] and bringing to the fore an interesting duality between these oracles, namely that they form a retract. Moving to syntax we show that Herbelin s language of classical Bohm trees from [Her97] defines any compact innocent strategy (in arenas interpreting simple types over integers) This second ....

....be described as rbt without #s. 4 Factorization We can rework the decomposition, but from within the model, in a more intrinsic way. We re only giving an informal argument that one can extrude case out of an innocent strategy and obtain something rigid, much in the same way as Laird proved in [Lai97] that one can extrude catch out of an innocent strategy and obtain something bracketed. To do this, and for reasons apparent in the proof, we have to restrict to arenas where Answers are terminal, i.e. condition (a4) Typical arenas where this is not the case are sums or lifted arenas as used in ....

J. Laird. Full abstraction for functional languages with control. In Proceedings, Twelfth Annual IEEE Symposium on Logic in Computer Science. IEEE Computer Society Press, 1997.

.... In the last few years, Game Semantics has provided a fully abstract model of Plotkin s functional calculus PCF [16, 15, 12, 4] and has been used to model a remarkably large range of programming language features, including recursive types [5] imperative variables [6] continuations [13], references [3] and nondeterminism [11] It is commonly believed that this success is at least in part due to the fact that Game Semantics explicitly considers concepts of an operational nature. Subtyping is a feature of type systems of programming languages which greatly facilitates the ....

....yields an interpretation of bounded quantification. Our construction is generic, in that it does not depend on the details of the calculus being used, and we believe that it could be adapted without trouble to more complex deterministic calculi for which Game models have been developed [6, 13]. Whether or not such extensions will yield any new types of interest, either from a logical or static analysis standpoint, remains to be seen. The most obvious shortcoming of this work is the failure of computational adequacy and of inequational soundness. The author believes that this is not an ....

J. Laird. Full abstraction for functional languages with control. In Proc. LICS'97, Warsaw, Poland, 1997.

.... A urry of remarkable theoretical results showed that game based techniques can give fully abstract models for pure functional languages [2,14,16] imperative languages with local variables and higher order procedures [3,4] languages with general references [1] languages with generalized control [15] and more. From a practical perspective, the game models have two attractive features. First, they are fully formal. The moves can be represented by elementary, uninterpreted, symbols; the rules are combinatorial in nature. This means that game models need not deal with the deep semantic issues of ....

Laird, J., Full abstraction for functional languages with control, in: Proceedings, Twelth Annual IEEE Symposium on Logic in Computer Science, IEEE Computer Society Press, Warsaw, Poland, 1997, pp. 58-67.

....x 1 ; x k in M terminates immediately when the term M tries to evaluate the variable x i and returns i Gamma 1. If M delivers n without using any of the x i , catch k x 1 ; x k in M returns n k. Thus one obtains the following semantic characterization of programming disciplines [AbrMcC97, Lai97, AbrHoMcC98, AbrMcC1] (the last case functional store control has not been published yet) Constraints Language D I B purely functional D I functional control D B functional store D functional store control Here D stands for the subcategory of C with the same objects and the morphisms restrained by ....

J. Laird, Full Abstraction for Functional Languages with Control, in: Proceedings of the Fourteenth International Symposium on Logic in Computer Science, Computer Society Press of the IEEE, 1997, p. 58--67

.... definability result [3, 17, 27] and hence provide the first syntax free constructions of fully abstract models of PCF [29] These results have since been extended to handle languages more complex than PCF, including untyped and recursively typed languages [5, 4] languages with control features [21], call by value languages [13, 6] and languages with side e#ects and store [7] In a slightly di#erent vein, games models have also been discovered for the constructive classical type theory [28] and for System F [14] Received July 11, 1997. The author acknowledges the support of Oxford ....

J. Laird, Full abstraction for functional languages with control, In Proceedings, Twelfth 362 GUY McCUSKER Annual IEEE Symposium on Logic in Computer Science [20], pp. 58--67.

....computations. The duality between them [19, 5] simply corresponds to the positive (CBV) negative (CBN) duality. The main direction for future work is the interpretation of programming languages (PCF, with control primitives. Such an interpretation has already been proposed by J. Laird [13], but is based on the HO game model for intuitionistic logic. An interpretation based on our model should be interesting, with the possibility of capturing both call by name and call by value in the same model. 1 Polarized games 1.1 De nitions We introduce our notion of polarized games by ....

James Laird. Full abstraction for functional languages with control. In Proceedings of the twelfth annual IEEE symposium on Logic In Computer Science, pages 58-67, Warsaw, June 1997. IEEE Computer Society Press.

.... cant point in this context is the close connection between the presented calculus and game semantics [3, 19, 22] the structure of interaction of typed processes with respect to (typed) environments precisely conforms to the intensional structures of games introduced in [22] and studied in e.g. [2, 10, 19, 24, 25]. It is notable that the type discipline itself does not mention basic notions in game semantics such as visibility, well bracketing and innocence (although it does use a syntactic form of IO alternation) yet they are derivable as operational properties of typed processes. We use this ....

Laird, J., Full abstraction for functional languages with control, LICS'97, IEEE, 1997. 16

....to give fully abstract models for many programming languages, including PCF [AJM96, HO96, Nic94] richer functional languages # Work supported by TMR Linear FMRX CT98 0170. 1 [AM95, McC96, HY97] and languages with non functional features such as reference types and non local control constructs [AM97, AM97a, AM97b, Lai97]. Once many concrete fully complete and fully abstract models have been studied, the problem of abstracting and axiomatizing the key properties of these constructions arises naturally. This line of research has been started by [Abr97] where axioms on models of PCF are given in order to guarantee ....

J.Laird. Full abstraction for functional languages with control, LICS Conf. Proc., 58--64, 1997.

....in Von Neumann matrix games. Indeed, the category of probabilistic strategies that we construct smoothly extends the deterministic games world as charted out by Hyland Ong [13] and Nickau [20] and further developed by Abramsky McCusker [3, 4, 5, 19] Abramsky, Honda McCusker [2] and Laird [15]. The category also fits in very nicely with the basic concepts of probability theory. The interaction of a strategy address: Equipe Preuves Programmes Syst emes, UFR de Math ematiques, Universit e de Paris 7, 2 place Jussieu, 75251 Paris Cedex 05, France. tel: 33 1 44 27 86 79. fax: 33 1 44 ....

J. Laird. Full abstraction for functional languages with control. In Proceedings, Twelfth Annual IEEE Symposium on Logic in Computer Science. IEEE Computer Society Press, 1997.

....to give fully abstract models for many programming languages, including PCF [AJM96, HO96, Nic94] richer functional languages Work supported by TMR Linear FMRX CT98 0170. 1 [AM95, McC96, HY97] and languages with non functional features such as reference types and non local control constructs [AM97, AM97a, AM97b, Lai97]. Once many concrete fully complete and fully abstract models have been studied, the problem of abstracting and axiomatizing the key properties of these constructions arises naturally. This line of research has been started by [Abr97] where axioms on models of PCF are given in order to guarantee ....

J.Laird. Full abstraction for functional languages with control, LICS Conf. Proc., 58--64, 1997.

.... models for various fragments of Linear Logic ( AJ94a,AM99] and to give fully abstract models for many programming languages, including PCF [AJM96,HO96,Nic94] richer functional languages [McC96] and languages with non functional features such as reference types and non local control constructs [AM97,Lai97]. Work partially supported by Linear FMRX CT98 0170. Once many concrete fully complete and fully abstract models have been studied, the problem of abstracting and axiomatizing the key properties of these constructions arises naturally. This line of research originated with [Abr97] where ....

J.Laird. Full abstraction for functional languages with control, LICS'97.

.... models for various fragments of Linear Logic ( AJ94a,AM99] and to give fully abstract models for many programming languages, including PCF [AJM96,HO96,Nic94] richer functional languages [McC96] and languages with non functional features such as reference types and non local control constructs [AM97,Lai97]. All these results are crucially based on the linear analysis of the intuitionistic arrow which is possible in the intensional setting of game categories. However, the definitions of game and game categories are quite complex, often requiring cumbersome quotienting operations. In this paper, we ....

J.Laird. Full abstraction for functional languages with control, LICS'97.

.... first syntax independent constructions of fully abstract models for a number of programming languages, including PCF [AJM96, HO96, Nic94] richer functional languages [AM95, McC96b, McC96a, HY97] and languages with non functional features such as reference types and non local control constructs [AM97c, AM97b, AM97a, Lai97]. A noteworthy feature is that the key definability results for the richer languages are proved by a reduction to definability for the functional fragment, using a technique of factorization theorems. Thus the results originally obtained for PCF prove to be a lynch pin in the analysis of a much ....

J. G. Laird. Full abstraction for functional languages with control. In Twelfth Annual IEEE Symposium on Logic in Computer Science, pages 58--64, 1997.

....(independent) semantic constraints on languages which were used to capture the purely functional model. In particular, local state is captured simply by allowing fully history sensitive strategies, while non local control is captured by relaxing the well bracketing constraint. A series of results [AM97a, AM97b, AM97c, Lai97, HY97] have shown that the correspondence between the syntactic and semantic cubes is indeed exact in the sense that the semantic models are fully abstract for the corresponding languages. A remarkable and unex3 pected feature of these results is that in each case definability for the model of the ....

J. Laird. Full abstraction for functional languages with control. In Proceedings of LiCS `97.

....obtained, showing that games models capture the behaviour of programs in a precise way. Furthermore, the paradigm is flexible enough to furnish fully abstract models of a wide variety of languages, ranging from purely functional languages [1, 4, 13, 16, 17, 19] to languages with control operators [15], mutable variables [3, 5] and even general references [2] However, despite being grounded on mathematical structures usually associated with concurrency, all of this work has focussed on sequential, deterministic languages. Contribution In this paper we present an account of nondeterminism in ....

J. Laird. Full abstraction for functional languages with control. In Proceedings, Twelth Annual IEEE Symposium on Logic in Computer Science [14].

....categorical universe in which representative functional calculi based on call by value evaluation can be faithfully interpreted. The independence of the parameter suggests we may obtain a suitable universe to model, say, imperative call by value computation by simply altering other parameters, cf. [4, 21]. We also note that the possibility to model data driven computation in contrast to demand driven computation as games is discussed in an early paper on game semantics by Abramsky and Jagadeesan [2] The main technical contribution of the present work is the validation of the semantic ....

....also be handled (though the premonoidal tensor in CBV poses a problem) as observed by Fiore and as will be reported elsewhere. For the interpretation of imperative constructs, we would consider, as noted in Introduction, variants of the present universe by changing parameters of games following [4, 21], which does lead to coherent semantic universes. One interesting topic in this context would be whether one needs refined type structures as in [4] for the interpretation of the impure constructs: indeed a much simpler, and more direct, approach seems possible in the present setting. Some results ....

Laird, J., Full abstraction for functional languages with control, LICS'97, IEEE, 1997.

....simple change of other parameters may result in the semantic universe of call by value computation capturing various language features. In the context of call by name computation, there are recent proposals to treat imperative constructs by Abramsky and McCusker [5] block structures) and by Laird [28] (control constructs) At least at the categorical level, the corresponding changes in the present setting result in coherent algebraic universes with distinct features, just by altering the concerned parameter. Indeed, just by changing the class of strategies following [5] we can simply ....

Laird, J., Full abstraction for functional languages with control, LICS'97, IEEE, 1997.

....for the interpretation of sums to follow. The latter is less fundamental for the results of the present paper, but in more recent work, Laird has fruitfully exploited the decoupling of the notion of view from the question answer structure to model languages with non local control operators [23]. A justified sequence s is well formed if it satisfies (w1) Players alternate: if s = s 1 mns 2 then OP (m) 6= OP (n) w2) The bracketing condition . We say that a question q in s is answered by a later answer a in s if q justifies a. The bracketing condition is satisfied by s if for each ....

.... the innocence condition gives a fully abstract model of Idealised Algol, a prototypical imperative language with procedures and local variables [6] while relaxing the bracketing condition, but retaining innocence, gives a fully abstract model of PCF extended with the control operator call cc [23]. This may also lead to a better understanding of the connections between game semantics and the intensional semantics provided by sequential algorithms [8] ....

J. Laird. Full abstraction for functional languages with control. Draft paper, December 1996.

....simple change of other parameters may result in the semantic universe of call by value computation capturing various language features. In the context of call by name computation, there are recent proposals to treat imperative constructs by Abramsky and McCusker [6] block structures) and by Laird [30] (control constructs) At least at the categorical level, the corresponding changes in the present setting result in coherent algebraic universes with distinct features, just by altering the concerned parameter. Indeed, just by changing the class of strategies following [6] we can simply ....

Laird, J., Full abstraction for functional languages with control, LICS'97, IEEE, 1997.

....of one of these constraints on strategies, leading to a larger category. Thus for example G b is the category of well bracketed but not necessarily innocent strategies. Remarkably, there is a precise correspondence between the syntactic and semantic squares, as shown in a series of papers [4,6,8,22]. For example, relaxing the constraint of innocence allows local state to be modelled, while relaxing bracketing allows control G i G G ib G b Figure 2: The semantic square operators to be modelled. Moreover, these increments in expressive power are exact, as shown by factorization theorems. ....

....We can now prove definability for the model of SPCF in the category C i . As was the case for IA, the proof makes use of a factorization theorem which removes violations of the bracketing condition from an arbitrary compact innocent strategy. The technique behind this factorization is due to Laird [22]. Lemma 39 Let A be an SPCF type and oe : A an innocently compact strategy. There exists a natural number k and an innocently compact, well bracketed strategy : N 1 Theta Delta Delta Delta Theta N k ) N k 1 ) N k 2 ) N such that c y k ; oe, where c k is the strategy used in ....

J. Laird. Full abstraction for functional languages with control. In Proceedings, Twelfth Annual IEEE Symposium on Logic in Computer Science, pages 58--67. IEEE Computer Society Press, 1997.

.... number of programming languages, including the prototypical functional language PCF [1, 8, 20] functional languages with richer type structures or different evaluation strategies [2,3,7,12,13] and languages with imperative features such as first order references and non local control constructs [4, 5, 11]. In the present paper, we obtain the first full abstraction result for a language with general references. By general references we mean references which can store not only values of ground types (integers, booleans, etc. but also those of higher types (procedures, higherorder functions, or ....

....location based models of store [15, 21, 25] or related denotational models of calculus [6, 24] It forms the basis of a precise semantic account of programs with general references, as shown by the full abstraction result. Technically, our results follow a similar pattern to those in [4, 5, 11], and exemplify the paradigm of the semantic cube proposed by the first author. In this paradigm, we consider firstly a syntactic cube of extended typed calculi. The origin of this cube is occupied by a purely functional language. Each axis of the cube corresponds to the extension of ....

[Article contains additional citation context not shown here]

J. Laird. Full abstraction for functional languages with control. In Proceedings, Twelfth Annual IEEE Symposium on Logic in Computer Science, pages 58--67. IEEE Computer Society Press, 1997.

No context found.

J. Laird. Full abstraction for functional languages with control. In Proceedings of the Twelfth International Symposium on Logic In Computer Science, LICS '97. IEEE Computer Society Press, 1997.

....theories are not consistent with the presence of exceptions. There are many other ways to model or reason about control, but one which deserves mention here is game semantics. The results described here arose from a semantic study of exceptions and continuations in a fully abstract games model [4, 5], although the presentation given here wholly eschews denotational semantics in order to make them more accessible. Thus one of the conclusions they support is a methodological one; game semantics with its focus on de nability and full abstraction can be a useful tool for investigating ....

J. Laird. Full abstraction for functional languages with control. In Proceedings of the Twelfth International Symposium on Logic In Computer Science, LICS '97. IEEE, 1997.

....entails that this operation is not e ective. On the other hand, the ne structure of Hyland Ong style games has allowed the creation of a hierarchy of models of calculi with nonfunctional features such as the references from Idealized Algol [2] and non determinism [15] control operators [20, 22] and higher type references [4] into which the intensionally fully abstract model of PCF can be embedded. Moreover, there are de nability results for each language, and each one except for PCF has an e ectively presentable fully abstract semantics obtainable by collapsing the games model under ....

....PCF, are the presence of error moves on the one hand, and the absence of question answer labelling of moves on the other. The reason that we have no need for the latter is that its sole function is to impose a bracketing condition or stack discipline for moves on legal sequences. As described in [20, 21], it is by abandoning this condition (in this case, by dropping the relevant labels) that non local control operators can be interpreted. Thus we have the following proposition from [20, 21] Proposition 3.20 There is a meaning preserving embedding of the intensionally fully abstract HO model of ....

[Article contains additional citation context not shown here]

J. Laird. Full abstraction for functional languages with control. In Proceedings of the Twelfth International Symposium on Logic In Computer Science, LICS '97. IEEE, 1997. 37

....intensional, combining ideas from concurrency theory and from traditional domain theory. A hierarchy of fully abstract models has emerged, based on the Hyland Ong (HO) model of PCF [12] extended with non functional (but still sequential) features such as mutable state and store [1,4] and control [13,14], to the point where there is now a thorough analysis of sequential functional computation which is to a large extent based on concurrency. Extending existing games models to give an interpretation of concurrent features is therefore a natural development, although it requires a substantial change ....

....ones (these results correspond to syntactic encodings of these features using threads and messages) Proposition 5.18 For every pointer blind strategy : A, there exists a wellbracketed (pointer blind) strategy b such that b . To prove this proposition, we use a result proved in [13] all violations of the bracketing condition can be expressed in terms of the single threaded strategy Peirce by factorization together with the fact that there is a well bracketed multi threaded strategy Peirce such that Peirce Peirce. Lemma 5.19 There is a strategy Peirce : nat ) ....

[Article contains additional citation context not shown here]

J. Laird. Full abstraction for functional languages with control. In Proceedings of the Twelfth International Symposium on Logic In Computer Science, LICS `97, 1997. 25 Laird

No context found.

J. Laird. Full abstraction for functional languages with control. Extended abstract, in the Proceedings of the 12th Annual Symposium on Logic in Computer Science, LICS '97, 1997.

No context found.

Laird, J., Full abstraction for functional languages with control, LICS'97, IEEE, 1997.

No context found.

J. Laird. Full abstraction for functional languages with control. In Twelveth IEEE Symposium on Logic in Computer Science, LiCS'97, pages 58--67, 1997.

No context found.

J. Laird, Full Abstraction for Functional Languages with Control, in: Proceedings of the Fourteenth International Symposium on Logic in Computer Science, Computer Society Press of the IEEE, 1997, p. 58--67

No context found.

Laird, J., Full abstraction for functional languages with control, LICS'97, IEEE, 1997. 16

No context found.

Laird, J., Full abstraction for functional languages with control, LICS'97, IEEE, 1997.

No context found.

James Laird. Full abstraction for functional languages with control. In Proc. LICS '97, 1997. (to appear).

No context found.

James Laird. Full abstraction for functional languages with control. In Proceedings of the Twelfth International Symposium on Logic in Computer Science (LICS'97), Warsaw, Poland, 1997.

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