leuven
play

@ Leuven H. Blockeel, J. Davis, L. De Raedt, D. Fierens, W. Meert, - PowerPoint PPT Presentation

Recent advances in lifted inference @ Leuven H. Blockeel, J. Davis, L. De Raedt, D. Fierens, W. Meert, N. Taghipour, G. Van den Broeck SML, April 19, 2012 Outline Introduction to lifted inference Four contributions Arbitrary


  1. Recent advances in lifted inference @ Leuven H. Blockeel, J. Davis, L. De Raedt, D. Fierens, W. Meert, N. Taghipour, G. Van den Broeck SML, April 19, 2012

  2. Outline  Introduction to lifted inference  Four contributions • Arbitrary constraints • Completeness results • Conditioning • An approximate method 1

  3. Lifted inference Exact Approximate Variable Knowledge … Belief … Elimination compilation propagation (2003) (2011) (2008) and many more ! 2

  4. 1.5 Attends(person) → Series MLN 1.2 Topic → Attends(person) 3

  5. 1.5 Attends(person) → Series MLN 1.2 Topic → Attends(person) Series Attends(p 1 ) Attends(p 2 ) … … Attends(p N ) Topic 3

  6. 1.5 Attends(person) → Series MLN 1.2 Topic → Attends(person) Series ϕ 1 ϕ 1 ϕ 1 ϕ 1 Attends(p 1 ) Attends(p 2 ) … … Attends(p N ) ϕ 2 ϕ 2 ϕ 2 ϕ 2 Topic 3

  7. 1.5 Attends(person) → Series MLN 1.2 Topic → Attends(person) Series ϕ 1 ϕ 1 ϕ 1 ϕ 1 Attends(p 1 ) Attends(p 2 ) … … Attends(p N ) ϕ 2 ϕ 2 ϕ 2 ϕ 2 Topic A 1 T ϕ 2 (A 1 ,T) true true 3.3 3.3 true false 1.0 false true false false 3.3 3

  8. 1.5 Attends(person) → Series MLN 1.2 Topic → Attends(person) Series ϕ 1 ϕ 1 ϕ 1 ϕ 1 Attends(p 1 ) Attends(p 2 ) … … Attends(p N ) ϕ 2 ϕ 2 ϕ 2 ϕ 2 Topic A 1 T ϕ 2 (A 1 ,T) A N T ϕ 2 (A N ,T) true true 3.3 true true 3.3 3.3 3.3 true false true false 1.0 1.0 false true false true false false false 3.3 false 3.3

  9. Series ϕ 1 ϕ 1 ϕ 1 ϕ 1 Attends(p 1 ) Attends(p 2 ) … … Attends(p N ) ϕ 2 ϕ 2 ϕ 2 ϕ 2 Topic 4

  10. Series ϕ 1 ϕ 1 ϕ 1 ϕ 1 Attends(p 1 ) Attends(p 2 ) … … Attends(p N ) ϕ 2 ϕ 2 ϕ 2 ϕ 2 Topic 1 N N   P ( S , A ,..., A , T ) ( A , S ) ( T , A )    1 N 1 i 2 i Z i 1 i 1   4

  11. Series ϕ 1 ϕ 1 ϕ 1 ϕ 1 Attends(p 1 ) Attends(p 2 ) … … Attends(p N ) ϕ 2 ϕ 2 ϕ 2 ϕ 2 Topic 1 N N     P ( S ) ... ( A , S ) ( T , A )    1 i 2 i Z T A A i 1 i 1   1 N will it become a series ? 4

  12. Series ϕ 1 ϕ 1 ϕ 1 ϕ 1 Attends(p 1 ) Attends(p 2 ) … … Attends(p N ) ϕ 2 ϕ 2 ϕ 2 ϕ 2 Topic 1 N N     P ( S ) ... ( A , S ) ( T , A )    1 i 2 i Z T A A i 1 i 1   1 N will it become 2 (N+1) terms a series ? 4

  13. Series ϕ 1 ϕ 1 ϕ 1 ϕ 1 Attends(p 1 ) Attends(p 2 ) … … Attends(p N ) ϕ 2 ϕ 2 ϕ 2 ϕ 2 Topic N N     ... ( A , S ) ( T , A )   1 i 2 i T A A i 1 i 1   1 N 2 (N+1) terms 4

  14. Series ϕ 1 ϕ 1 ϕ 1 ϕ 1 Attends(p 1 ) Attends(p 2 ) … … Attends(p N ) ϕ 2 ϕ 2 ϕ 2 ϕ 2 Topic            ( A , S ) ( T , A ) ... ( A , S ) ( T , A )         1 1 2 1 1 N 2 N     T A A 1 N 1 for every person 4

  15. Series ϕ 1 ϕ 1 ϕ 1 ϕ 1 Attends(p 1 ) Attends(p 2 ) … … Attends(p N ) ϕ 2 ϕ 2 ϕ 2 ϕ 2 Topic            ( A , S ) ( T , A ) ... ( A , S ) ( T , A )         1 1 2 1 1 N 2 N     T A A 1 N N times the same product ! 4

  16. Series ϕ 1 ϕ 1 ϕ 1 ϕ 1 Attends(p 1 ) Attends(p 2 ) … … Attends(p N ) ϕ 2 ϕ 2 ϕ 2 ϕ 2 Topic            ( A , S ) ( T , A ) ... ( A , S ) ( T , A )         1 1 2 1 1 N 2 N     T A A 1 N N times the same sum ! 4

  17. Series ϕ 1 ϕ 1 ϕ 1 ϕ 1 Attends(p 1 ) Attends(p 2 ) … … Attends(p N ) ϕ 2 ϕ 2 ϕ 2 ϕ 2 Topic N     ( A , S ) ( T , A )     lifted: 1 2   T A compute only once ! 4

  18. Series ϕ 1 ϕ 1 ϕ 1 ϕ 1 Attends(p 1 ) Attends(p 2 ) … … Attends(p N ) ϕ 2 ϕ 2 ϕ 2 ϕ 2 Topic N     ( A , S ) ( T , A )     lifted: 1 2   T A “lifted multiplication” “lifted sum - out” 4

  19. Lifted Variable Elimination [Poole ’03,…]  Repeatedly apply certain operators on the model • Lifted multiplication • Lifted sum-out • …  Until the desired result is found 5

  20. Lifted Knowledge Compilation [Van den Broeck et al ‘11,…]  Compile the model into a “lifted” circuit (“FO d - DNNF”) • How? Compilation rules  Inference = traversing the circuit • Time = poly(domain size) 8

  21. Outline  Introduction to lifted inference  Four contributions • Arbitrary constraints • Completeness results • Conditioning • An approximate method 9

  22. S A(p 1 ) … A(p N/2 ) A(p N/2+1 ) … A(p N ) T 10

  23. S A(p 1 ) … A(p N/2 ) A(p N/2+1 ) … A(p N ) T 10

  24. S A(p 1 ) … A(p N/2 ) A(p N/2+1 ) … A(p N ) T 10

  25. S A(p 1 ) … A(p N/2 ) A(p N/2+1 ) … A(p N ) T 10

  26. S A(p 1 ) … A(p N/2 ) A(p N/2+1 ) … A(p N ) T 10

  27. S A(p 1 ) … A(p N/2 ) A(p N/2+1 ) … A(p N ) T 10

  28. S A(p 1 ) … A(p N/2 ) A(p N/2+1 ) … A(p N ) T Bigger groups = more lifting ! 10

  29. S A(p 1 ) … A(p N/2 ) A(p N/2+1 ) … A(p N ) T Bigger groups = more lifting ! The groups are specified by constraints 10

  30. Importance of constraints [Taghipour et al, AISTATS'12]  Exact lifted algorithms use a particular constraint language can it be expressed group → constraint → in the language ?  Often leads to unnecessarily small groups → less lifting 11

  31. Importance of constraints [Taghipour et al, AISTATS'12]  Exact lifted algorithms use a particular constraint language can it be expressed group → constraint → in the language ?  Often leads to unnecessarily small groups → less lifting  We avoid using a particular constraint language Instead: arbitrary constraints + relational algebra 11

  32. pairwise constraints (C-FOVE) runtime (log) arbitrary constraints more evidence 12

  33. Outline  Introduction to lifted inference  Four contributions • Arbitrary constraints • Completeness results • Conditioning • An approximate method 13

  34. Outline ● Introduction to lifted inference ● Four contributions ● Arbitrary constraints ● Completeness result ● Conditioning ● Approximate inference

  35. Outline ● Introduction to lifted inference ● Four contributions ● Arbitrary constraints ● Completeness result ● Conditioning ● Approximate inference

  36. What is Lifted Inference? ● Propositional inference is intractable Solution: lifted inference “Exploit symmetries” “Reason at first-order level” “Reason about groups of objects as a whole” “Avoid repeated computations” “Mimic resolution in theorem proving” ● There is a common understanding but no formal definition of lifted inference!

  37. What is Lifted Inference? ● What is commonly understood as exact lifted inference? Definition: Domain-Lifted Inference Complexity of computing P( q | e ) in model m is polynomial time in the domain sizes of the logical variables in q,e,m 1.5 Attends(person) → Series 1.2 Topic → Attends(person) [Van den Broeck NIPS11]

  38. What is Lifted Inference? ● What is commonly understood as exact lifted inference? Definition: Domain-Lifted Inference Complexity of computing P( q | e ) in model m is polynomial time in the domain sizes of the logical variables in q,e,m ● Possibly exponential in the size of q,e,m # predicates, # parfactors, # atoms, # arguments, # formulas, # constants in model [Van den Broeck NIPS11]

  39. What is Lifted Inference? ● Motivation: Large domains lead to intractable propositional inference. ● A formal framework for lifted inference ● Definition + complexity considerations ● ~ PAC-learnability (Valiant) ● Other notions, e.g., for approximate inference. [Van den Broeck NIPS11]

  40. Completeness ● A procedure that is domain-lifted for all models in a class M is called complete for M All models in M are “liftable” ● There was no completeness result for existing algorithms If you give me a model, I cannot say if grounding will be needed, untill I run the inference algorithm itself. [Van den Broeck NIPS11]

  41. Completeness Result Probabilistic inference in models with ● universal quantifiers ∀ and ● 2 logical variables per formula is domain-liftable. ● A non-trivial class of models ● First completeness results in exact lifted inference ● Lifted knowledge compilation procedure ● Lifted variable elimination procedure [Van den Broeck NIPS11], [Taghipour et al.]

  42. Completeness Game No domain-lifted inference procedure exists FOL , ,= ∀ ∃ [Jaeger 99] Expressivity ... [Jaeger 12] ? ∀ =, 2 variables [Van den Broeck 11] FOL , Complete domain-lifted inference procedure

  43. Outline ● Introduction to lifted inference ● Four contributions ● Arbitrary constraints ● Completeness result ● Conditioning ● Approximate inference

  44. Conditioning ● Task: Probability of query q given evidence e : P( q | e ) Domain-lifted inference is exponential in the size of e . ● Can we compute conditional probabilities efficiently? Depends on the arity of literals conditioned on: ● Positive and negative result for lifted inference [Van den Broeck, Davis AAAI12]

Recommend


More recommend