1 / 20

Rules Dependencies in Backward Chaining of Conceptual Graphs Rules

Rules Dependencies in Backward Chaining of Conceptual Graphs Rules. Jean-Franç ø is B å get LIRMM / INRIA Rhône-Alpes jean-francois.baget@inrialpes.fr Eric S å lv å t IMERIR salvat@imerir.com. Context: optimization of deduction with CG Rules. Deduction in SG [Sowa:76]

bat
Télécharger la présentation

Rules Dependencies in Backward Chaining of Conceptual Graphs Rules

An Image/Link below is provided (as is) to download presentation Download Policy: Content on the Website is provided to you AS IS for your information and personal use and may not be sold / licensed / shared on other websites without getting consent from its author. Content is provided to you AS IS for your information and personal use only. Download presentation by click this link. While downloading, if for some reason you are not able to download a presentation, the publisher may have deleted the file from their server. During download, if you can't get a presentation, the file might be deleted by the publisher.

E N D

Presentation Transcript


  1. Rules Dependencies in Backward Chaining of Conceptual Graphs Rules Jean-Françøis Båget LIRMM / INRIA Rhône-Alpes jean-francois.baget@inrialpes.fr Eric Sålvåt IMERIR salvat@imerir.com

  2. Context: optimization of deduction with CG Rules • Deduction in SG[Sowa:76] • Deduction in SR[Sowa:84;Salvat,Mugnier:96] • Optimizing deduction in SR : piece unification for Backward Chaining [Salvat, Mugnier:96; Salvat:98] • Optimizing deduction in SR : rules dependencies for Forward Chaining [Baget: 04] 30 !

  3. Caveat • J.-François/Eric paper := (Intro . (Definition | Theorem | Proof)* . Concl) • This time it’s even worse: • No example ! • No graph drawing ! • No poetry …

  4. hyp conc Caveat IF Agnt subject Person: JFB Present Document CG Obj THEN member G Document CG Team: RCR Paper contains subject interest subject contains Drawing CG Drawing subject

  5. Simple Graph G Simple Graph H  G V H   , Overview of deduction in SG Vocabulary V

  6. Projection: a deduction calculus in SG V PROJECTION? is a NP-complete problem H G  Theorem [Sowa:84; Chein, Mugnier:96]: G V H iff there is a projection  from H into (the normal form of) G.

  7. G, RV H Set of Rules R con hyp con hyp   , , , Overview of deduction in SR Vocabulary V Simple Graph H Simple Graph G   

  8. Forward Chaining: a deduction calculus in SR V con hyp con hyp R  Theorem [Salvat, Mugnier:96; Salvat:98]: G, RV H iff there exists k s.t. H projects into [R]k(G).  ? G H (normal form) [R]1(G) (normal form) [R]2(G)

  9. (Un)decidability of deduction in SR [R]2(G) [R]1(G) [R]k(G) V[R]k+1(G) • Examples of finite expansion sets • Disconnected rules • Range restricted rules • The union of 2 f.e.s. is not necessarily a f.e.s. Definition [Baget, Mugnier:02]: R is a finite expansion set iff  G,  k / [R]k(G) V [R]k+1(G)  ? G Theorem [Coulondre, Salvat:98]: Deduction in SR is undecidable (semi-decidable). H

  10. Dependencies between rules Definition [Baget:04]: A rule R1depends upon a rule R2 iff there exists a graph G such that applying R2 on G creates a new application of R1. R1 R2 Precompilation of dependencies reduces the number of applicability tests in Forward Chaining… DEPENDS? is a NP-complete problem G G [R]1(G) Suppose now that R1 does not depend upon R2, and use Forward Chaining…

  11. Graph of rules dependencies (GRD) [R]3(G) 1 6 4 G 2 [R]1(G) 5 [R]2(G) 3 [R]4(G) R Theorem [Baget:04]: Deduction in SR is decidable when the GRD contains no circuit. N2 calls to a NP-hard probem …

  12. Graph of rules dependencies (GRD) Disconnected rules 1 6 4 G 2 [R]1(G) 5 3 [R]k(G) R Theorem [Baget:04]: Deduction in SR is decidable when all strongly connected components of the GRD are f.e.s. [R]k’(G) Range-restricted rule [R]k’+1(G)

  13. Graph of rules dependencies (GRD) 1 6 4 G  2 5 7 3 R 8 H 

  14. Using proofs of dependencies R1 R2 • is a linear time operator    G Theorem [Baget:04]: If ’ is a new projection from hyp(R2) into G’, then ’ extends   G’

  15. Backward Chaining: a deduction calculus in SR H hyp(R) H’ Piece unification [Salvat:98] R

  16. Backward Chaining: a deduction calculus in SR  con hyp G G con hyp R Theorem [Salvat, Mugnier:96; Salvat:98]: G, RV H iff there exists a sequence of piece unifications that transforms H into the empty SG. H H H H

  17. So, what’s new in this paper ? • Different representations • Hypergraphs, colored graphs [Baget:04] • Multigraphs, lambda abtractions [Salvat:98] • Different restrictions • Lattice as concept types hierarchy [Salvat:98] • Poor treatment of individuals in conclusion [Baget:04] • Improving both results • Unification of syntaxes • Removal of all restrictions • Extension to conjunctive concept types (collateral benefit) =

  18. Using the GRD in Backward Chaining 1 6 4 G  2 Theorem [Baget, Salvat:06]: H’ can only be unified with predecessors of H or predecessors of the rule used to obtain H’. 5 7 3 R 8 H H’

  19. Using the GRD in Backward Chaining • Reduces the # of rules used in BC • as in FC, remove rules that are not on a path from G to H • Reduces # of unification checks in BC • as in FC, only checks for neighbours in the GRD • Reduces the cost of unification checks ? • in FC,   linearly computes partial projections to extend. • In BC, we should obtain a partial unification  to extend ….

  20. Thanks for your attention Applause Thank you … Questions

More Related