recursive conditioning adnan darwiche computer science department ucla
DESCRIPTION
Recursive Conditioning Adnan Darwiche Computer Science Department UCLA. Decomposition. Battery Age. Alternator. Fan Belt. Leak. Charge Delivered. Battery. Fuel Line. Starter. Gas. Distributor. Battery Power. Spark Plugs. Gas Gauge. Engine Start. Lights. Engine Turn Over. Radio. - PowerPoint PPT PresentationTRANSCRIPT
![Page 1: Recursive Conditioning Adnan Darwiche Computer Science Department UCLA](https://reader034.vdocument.in/reader034/viewer/2022052604/568143cf550346895db05cdc/html5/thumbnails/1.jpg)
Recursive Conditioning
Adnan DarwicheComputer Science Department
UCLA
![Page 2: Recursive Conditioning Adnan Darwiche Computer Science Department UCLA](https://reader034.vdocument.in/reader034/viewer/2022052604/568143cf550346895db05cdc/html5/thumbnails/2.jpg)
Decomposition
Battery Age Alternator Fan Belt
BatteryCharge Delivered
Battery Power
Starter
Radio Lights Engine Turn Over
Gas Gauge
Gas
Leak
Fuel Line
Distributor
Spark Plugs
Engine Start
![Page 3: Recursive Conditioning Adnan Darwiche Computer Science Department UCLA](https://reader034.vdocument.in/reader034/viewer/2022052604/568143cf550346895db05cdc/html5/thumbnails/3.jpg)
Decomposition
Battery Age Alternator Fan Belt
BatteryCharge Delivered
Battery Power
Starter
Radio Lights Engine Turn Over
Gas Gauge
Gas
Leak
Fuel Line
Distributor
Spark Plugs
Engine Start
![Page 4: Recursive Conditioning Adnan Darwiche Computer Science Department UCLA](https://reader034.vdocument.in/reader034/viewer/2022052604/568143cf550346895db05cdc/html5/thumbnails/4.jpg)
Decomposition
Battery Age Alternator Fan Belt
BatteryCharge Delivered
Battery Power
Starter
Radio Lights Engine Turn Over
Gas Gauge
Gas
Leak
Fuel Line
Distributor
Spark Plugs
Engine Start
![Page 5: Recursive Conditioning Adnan Darwiche Computer Science Department UCLA](https://reader034.vdocument.in/reader034/viewer/2022052604/568143cf550346895db05cdc/html5/thumbnails/5.jpg)
DecompositionBattery Age Alternator Fan Belt
BatteryCharge Delivered
Battery Power
Starter
Radio Lights Engine Turn Over
Gas Gauge
Gas
Leak
Fuel Line
Distributor
Spark Plugs
Engine Start
LP RP*
![Page 6: Recursive Conditioning Adnan Darwiche Computer Science Department UCLA](https://reader034.vdocument.in/reader034/viewer/2022052604/568143cf550346895db05cdc/html5/thumbnails/6.jpg)
DecompositionBattery Age Alternator Fan Belt
BatteryCharge Delivered
Battery Power
Starter
Radio Lights Engine Turn Over
Gas Gauge
Gas
Leak
Fuel Line
Distributor
Spark Plugs
Engine Start
LP RP*
![Page 7: Recursive Conditioning Adnan Darwiche Computer Science Department UCLA](https://reader034.vdocument.in/reader034/viewer/2022052604/568143cf550346895db05cdc/html5/thumbnails/7.jpg)
Causal Network
Battery Age Alternator Fan Belt
BatteryCharge Delivered
Battery Power
Starter
Radio Lights Engine Turn Over
Gas Gauge
Gas
Leak
Fuel Line
Distributor
Spark Plugs
Engine Start
![Page 8: Recursive Conditioning Adnan Darwiche Computer Science Department UCLA](https://reader034.vdocument.in/reader034/viewer/2022052604/568143cf550346895db05cdc/html5/thumbnails/8.jpg)
Causal Network
Battery Age Alternator Fan Belt
BatteryCharge Delivered
Battery Power
Starter
Radio Lights Engine Turn Over
Gas Gauge
Gas
Leak
Fuel Line
Distributor
Spark Plugs
Engine Start
![Page 9: Recursive Conditioning Adnan Darwiche Computer Science Department UCLA](https://reader034.vdocument.in/reader034/viewer/2022052604/568143cf550346895db05cdc/html5/thumbnails/9.jpg)
Case AnalysisBattery Age Alternator Fan Belt
Battery
Charge Delivered
Battery Power
Starter
Radio Lights Engine Turn Over
Gas Gauge
Gas
Leak
Fuel Line
Distributor
Spark Plugs
Engine Start
Case I
Battery Age Alternator Fan Belt
Battery
Charge Delivered
Battery Power
Starter
Radio Lights Engine Turn Over
Gas Gauge
Gas
Leak
Fuel Line
Distributor
Spark Plugs
Engine Start
Case II
![Page 10: Recursive Conditioning Adnan Darwiche Computer Science Department UCLA](https://reader034.vdocument.in/reader034/viewer/2022052604/568143cf550346895db05cdc/html5/thumbnails/10.jpg)
Case AnalysisBattery Age Alternator Fan Belt
Battery
Charge Delivered
Battery Power
Starter
Radio Lights Engine Turn Over
Gas Gauge
Gas
Leak
Fuel Line
Distributor
Spark Plugs
Engine Start
Battery Age Alternator Fan Belt
Battery
Charge Delivered
Battery Power
Starter
Radio Lights Engine Turn Over
Gas Gauge
Gas
Leak
Fuel Line
Distributor
Spark Plugs
Engine Start
Case I Case II
LP * RP
![Page 11: Recursive Conditioning Adnan Darwiche Computer Science Department UCLA](https://reader034.vdocument.in/reader034/viewer/2022052604/568143cf550346895db05cdc/html5/thumbnails/11.jpg)
Case AnalysisBattery Age Alternator Fan Belt
Battery
Charge Delivered
Battery Power
Starter
Radio Lights Engine Turn Over
Gas Gauge
Gas
Leak
Fuel Line
Distributor
Spark Plugs
Engine Start
Battery Age Alternator Fan Belt
Battery
Charge Delivered
Battery Power
Starter
Radio Lights Engine Turn Over
Gas Gauge
Gas
Leak
Fuel Line
Distributor
Spark Plugs
Engine Start
Case I Case II
LP * RP LP * RP+
![Page 12: Recursive Conditioning Adnan Darwiche Computer Science Department UCLA](https://reader034.vdocument.in/reader034/viewer/2022052604/568143cf550346895db05cdc/html5/thumbnails/12.jpg)
Case AnalysisBattery Age Alternator Fan Belt
BatteryCharge Delivered
Battery Power
Starter
Radio Lights Engine Turn Over
Gas Gauge
Gas
Leak
Fuel Line
Distributor
Spark Plugs
Engine Start
LP * RP LP * RP+
![Page 13: Recursive Conditioning Adnan Darwiche Computer Science Department UCLA](https://reader034.vdocument.in/reader034/viewer/2022052604/568143cf550346895db05cdc/html5/thumbnails/13.jpg)
Battery Age Alternator Fan Belt
Battery
Charge Delivered
Battery Power
Starter
Radio Lights Engine Turn Over
Gas Gauge
Gas
Leak
Fuel Line
Distributor
Spark Plugs
Engine Start
Battery Age Alternator Fan Belt
Battery
Charge Delivered
Battery Power
Starter
Radio Lights Engine Turn Over
Gas Gauge
Gas
Leak
Fuel Line
Distributor
Spark Plugs
Engine Start
• Decomposition and Case Analysis can answer any query
• Non-Deterministic!
![Page 14: Recursive Conditioning Adnan Darwiche Computer Science Department UCLA](https://reader034.vdocument.in/reader034/viewer/2022052604/568143cf550346895db05cdc/html5/thumbnails/14.jpg)
Decomposition Tree
A B C D E
A A B B C
C DDB
E
B
![Page 15: Recursive Conditioning Adnan Darwiche Computer Science Department UCLA](https://reader034.vdocument.in/reader034/viewer/2022052604/568143cf550346895db05cdc/html5/thumbnails/15.jpg)
Decomposition Tree
A B C D E
A A B B C
C DDB
E
B
![Page 16: Recursive Conditioning Adnan Darwiche Computer Science Department UCLA](https://reader034.vdocument.in/reader034/viewer/2022052604/568143cf550346895db05cdc/html5/thumbnails/16.jpg)
Decomposition Tree
A B C D E
A A B C
C DD E
B
![Page 17: Recursive Conditioning Adnan Darwiche Computer Science Department UCLA](https://reader034.vdocument.in/reader034/viewer/2022052604/568143cf550346895db05cdc/html5/thumbnails/17.jpg)
Decomposition Tree
A B C D E
A A B B C
C DDB
E
B
![Page 18: Recursive Conditioning Adnan Darwiche Computer Science Department UCLA](https://reader034.vdocument.in/reader034/viewer/2022052604/568143cf550346895db05cdc/html5/thumbnails/18.jpg)
Decomposition Tree
A B C D E
A A B B C
C DDB
E
B
![Page 19: Recursive Conditioning Adnan Darwiche Computer Science Department UCLA](https://reader034.vdocument.in/reader034/viewer/2022052604/568143cf550346895db05cdc/html5/thumbnails/19.jpg)
Decomposition Tree
A B C D E
A A B C
C DD E
BLP RP
![Page 20: Recursive Conditioning Adnan Darwiche Computer Science Department UCLA](https://reader034.vdocument.in/reader034/viewer/2022052604/568143cf550346895db05cdc/html5/thumbnails/20.jpg)
Decomposition Tree
A B C D E
A A B B C
C DDB
E
B LP * RP
![Page 21: Recursive Conditioning Adnan Darwiche Computer Science Department UCLA](https://reader034.vdocument.in/reader034/viewer/2022052604/568143cf550346895db05cdc/html5/thumbnails/21.jpg)
Decomposition Tree
A B C D E
A A B B C
C DDB
E
B LP * RP
![Page 22: Recursive Conditioning Adnan Darwiche Computer Science Department UCLA](https://reader034.vdocument.in/reader034/viewer/2022052604/568143cf550346895db05cdc/html5/thumbnails/22.jpg)
Decomposition Tree
A B C D E
A A B C
C DD E
B LP * RP
LP RP
![Page 23: Recursive Conditioning Adnan Darwiche Computer Science Department UCLA](https://reader034.vdocument.in/reader034/viewer/2022052604/568143cf550346895db05cdc/html5/thumbnails/23.jpg)
Decomposition Tree
A B C D E
A A B B C
C DDB
E
B LP * RP LP * RP
![Page 24: Recursive Conditioning Adnan Darwiche Computer Science Department UCLA](https://reader034.vdocument.in/reader034/viewer/2022052604/568143cf550346895db05cdc/html5/thumbnails/24.jpg)
Decomposition Tree
A B C D E
A A B B C
C DDB
E
LP * RP + LP * RP
![Page 25: Recursive Conditioning Adnan Darwiche Computer Science Department UCLA](https://reader034.vdocument.in/reader034/viewer/2022052604/568143cf550346895db05cdc/html5/thumbnails/25.jpg)
Dtrees
Generation from– Elimination orders– Jointree– hMeTiS hypergraph
partitioning program
a a’
.60 .40
b b’
a .90 .10
a’ .20 .80
d d’
c .50 .50
c’ .40 .60
e e’
bd .25 .75
bd’ .25 .75
b’d .90 .10
b’d’ .05 .95
c c’
b .70 .30
b’ .15 .85
A B C DE
A C
B
D
• CutsetB B
BC
• Context• Cluster
• Vars
A AB
AB BCDE
BC
CD BDE
BCDE
![Page 26: Recursive Conditioning Adnan Darwiche Computer Science Department UCLA](https://reader034.vdocument.in/reader034/viewer/2022052604/568143cf550346895db05cdc/html5/thumbnails/26.jpg)
RC1
RC1(T,e)if T is a leaf nodereturn Lookup(T,e)else
p := 0for each instantiation c of cutset(T)-E do
p := p + RC1(Tl,ec) RC1(Tr,ec)
return p
![Page 27: Recursive Conditioning Adnan Darwiche Computer Science Department UCLA](https://reader034.vdocument.in/reader034/viewer/2022052604/568143cf550346895db05cdc/html5/thumbnails/27.jpg)
Lookup(T,e)X|U : CPT associated with leaf T
If X is instantiated in e, thenx: value of X in eu: value of U in eReturn x|u
Else return 1 = xx|u
![Page 28: Recursive Conditioning Adnan Darwiche Computer Science Department UCLA](https://reader034.vdocument.in/reader034/viewer/2022052604/568143cf550346895db05cdc/html5/thumbnails/28.jpg)
Computational Complexity
A B C D E
A A B B C
C DDB
E
A
B
C
D
B
BC
BCDAncestoral Cutset
![Page 29: Recursive Conditioning Adnan Darwiche Computer Science Department UCLA](https://reader034.vdocument.in/reader034/viewer/2022052604/568143cf550346895db05cdc/html5/thumbnails/29.jpg)
Computational Complexity• Given
– DAG with n nodes– elimination order of width w
• Can construct a dtree in O(n log n) time:– Height O(log n)– cutset width <= w+1– a-cutset width O(w log n)
• Time complexity: O(n cw log n)• Cutset Conditioning: O(n c
s)
![Page 30: Recursive Conditioning Adnan Darwiche Computer Science Department UCLA](https://reader034.vdocument.in/reader034/viewer/2022052604/568143cf550346895db05cdc/html5/thumbnails/30.jpg)
Network Effective Network
Size
Elimination-Order Width
Loop-Cutset Width
A-Cutset Width
1 Water 59.0 21.3 29.5 32.32 Midlew 108.4 19.7 39.3 45.93 Barley 138.0 21.8 57.3 51.14 Diabetes 1349.1 19.2 557.2 77.95 Link 922.7 28.0 347.0 70.76 Pigs 699.0 16.4 144.2 38.07 Munin1 410.3 26.4 122.6 51.78 Munin2 2202.7 20.0 495.9 54.69 Munin3 2293.4 16.3 454.2 51.1
10 Munin4 2313.4 19.7 521.5 51.9
![Page 31: Recursive Conditioning Adnan Darwiche Computer Science Department UCLA](https://reader034.vdocument.in/reader034/viewer/2022052604/568143cf550346895db05cdc/html5/thumbnails/31.jpg)
Relation to Jointrees
Time: O(n cw)Space: O(n)
Time: O(n cwlog n)
Space: O(n cw)
![Page 32: Recursive Conditioning Adnan Darwiche Computer Science Department UCLA](https://reader034.vdocument.in/reader034/viewer/2022052604/568143cf550346895db05cdc/html5/thumbnails/32.jpg)
Decomposition Tree
A B C D E F
A
A B
B C
C D
D E E F
AB
CA B C
![Page 33: Recursive Conditioning Adnan Darwiche Computer Science Department UCLA](https://reader034.vdocument.in/reader034/viewer/2022052604/568143cf550346895db05cdc/html5/thumbnails/33.jpg)
Decomposition Tree
A B C D E F
A
A B
B C
C D
D E E F
AB
C CC
.27
.39
ABCABCABCABCABCABCABCABC
A B C
Context(N)= A-Cutset(N)&Vars(N)
![Page 34: Recursive Conditioning Adnan Darwiche Computer Science Department UCLA](https://reader034.vdocument.in/reader034/viewer/2022052604/568143cf550346895db05cdc/html5/thumbnails/34.jpg)
Decomposition Tree
A B C D E F
A
A B
B C
C D
D E E F
AB
CD
E
AB
CD
![Page 35: Recursive Conditioning Adnan Darwiche Computer Science Department UCLA](https://reader034.vdocument.in/reader034/viewer/2022052604/568143cf550346895db05cdc/html5/thumbnails/35.jpg)
RC2
RC2(T,e)if T is a leaf node, return Lookup(T,e)y := instantiation of context(T)If cacheT[y] <> nil, return cacheT[y]
p := 0for each instantiation c of cutset(T)-E do
p := p + RC2(Tl,ec) RC2(Tr,ec)
cacheT[y] := p
return p
![Page 36: Recursive Conditioning Adnan Darwiche Computer Science Department UCLA](https://reader034.vdocument.in/reader034/viewer/2022052604/568143cf550346895db05cdc/html5/thumbnails/36.jpg)
Constructing dtreesA
DB
E
C
F
F E A B C D
A AB AC ABE BCD DF
DBC
A
W=2
Cutset <= w+1
![Page 37: Recursive Conditioning Adnan Darwiche Computer Science Department UCLA](https://reader034.vdocument.in/reader034/viewer/2022052604/568143cf550346895db05cdc/html5/thumbnails/37.jpg)
Constructing dtreesA
DB
E
C
F
F E A B C D
A AB AC ABE BCD DF
DBC
A
W=2
![Page 38: Recursive Conditioning Adnan Darwiche Computer Science Department UCLA](https://reader034.vdocument.in/reader034/viewer/2022052604/568143cf550346895db05cdc/html5/thumbnails/38.jpg)
Constructing dtreesA
D
B
E
C
F
F E A B C D
A AB AC ABE BCD DF
DBC
A
W=2
![Page 39: Recursive Conditioning Adnan Darwiche Computer Science Department UCLA](https://reader034.vdocument.in/reader034/viewer/2022052604/568143cf550346895db05cdc/html5/thumbnails/39.jpg)
Constructing dtreesA
D
B
E
C
F
F E A B C D
A AB AC ABE BCD DF
DBC
A
W=2
![Page 40: Recursive Conditioning Adnan Darwiche Computer Science Department UCLA](https://reader034.vdocument.in/reader034/viewer/2022052604/568143cf550346895db05cdc/html5/thumbnails/40.jpg)
Constructing dtreesA
DB
E
C
F
F E A B C D
A AB AC ABE BCD DF
DBC
A
W=2
DBC
AB ABC
Context <= w+1
Space: O(n cw)
![Page 41: Recursive Conditioning Adnan Darwiche Computer Science Department UCLA](https://reader034.vdocument.in/reader034/viewer/2022052604/568143cf550346895db05cdc/html5/thumbnails/41.jpg)
Constructing dtreesA
DB
E
C
F
F E A B C D
A AB AC ABE BCD DF
DBCD
ABC
W=2
AB ABC
Cluster<= w+1
Space: O(n cw)Time: O(n cw)
![Page 42: Recursive Conditioning Adnan Darwiche Computer Science Department UCLA](https://reader034.vdocument.in/reader034/viewer/2022052604/568143cf550346895db05cdc/html5/thumbnails/42.jpg)
Constructing dtreesA
DB
E
C
F
F E A B C D
ABE DF
BCDABC
W=2
Cluster<= w+1
![Page 43: Recursive Conditioning Adnan Darwiche Computer Science Department UCLA](https://reader034.vdocument.in/reader034/viewer/2022052604/568143cf550346895db05cdc/html5/thumbnails/43.jpg)
Graphical Models
Elimination Order Jointree
Dtree
![Page 44: Recursive Conditioning Adnan Darwiche Computer Science Department UCLA](https://reader034.vdocument.in/reader034/viewer/2022052604/568143cf550346895db05cdc/html5/thumbnails/44.jpg)
Hypergraph partitioningHypergraph partitioning
The problem of hypergraphpartitioning is well-studied
in VLSI design.….and is alive!
A hypergraph is a generalization of a graph, such that an edge is permitted to connect an arbitrary number of vertices, rather than exactly two.
The task of hypergraph partitioning is to find a way to split the vertices of a hypergraph into k approximately equal parts, such that the number of hyperedges connecting vertices in different parts is minimized.
![Page 45: Recursive Conditioning Adnan Darwiche Computer Science Department UCLA](https://reader034.vdocument.in/reader034/viewer/2022052604/568143cf550346895db05cdc/html5/thumbnails/45.jpg)
An algorithm using hypergraph An algorithm using hypergraph partitioning to construct dtreespartitioning to construct dtrees
Given a DAG G, we first generate a hypergraph H that corresponds to G:
For each family F in G, we add a node NF to H. For each variable V in G, we add a hyperedge to H which connects all nodes NF such that V is a member of F.
Then we recursively invoke hypergraph partitioning to build a dtree:
The algorithm partitions the hypergraph into two sets of vertices, then recursively generates dtrees for each set, and finally combines the resulting dtrees into a new dtree (whose left child is the dtree for the first set, and whose right child is the dtree for the second set).
A AB
CDBC
A AB CDBC
A AB CDBC
For DAG:
![Page 46: Recursive Conditioning Adnan Darwiche Computer Science Department UCLA](https://reader034.vdocument.in/reader034/viewer/2022052604/568143cf550346895db05cdc/html5/thumbnails/46.jpg)
A
CB
ED
F
HG
DFDF
AC ACE
EFH A AF
AB ABD
DFG
CC
EEHH
AAGG
BB
From dtrees to elimination ordersFrom dtrees to elimination orders
![Page 47: Recursive Conditioning Adnan Darwiche Computer Science Department UCLA](https://reader034.vdocument.in/reader034/viewer/2022052604/568143cf550346895db05cdc/html5/thumbnails/47.jpg)
Any-Space Inference
![Page 48: Recursive Conditioning Adnan Darwiche Computer Science Department UCLA](https://reader034.vdocument.in/reader034/viewer/2022052604/568143cf550346895db05cdc/html5/thumbnails/48.jpg)
32
8 16
864 128
1024 512
1728 cache entries 64 cache entries
Time-Space TradeoffsTime-Space Tradeoffs
![Page 49: Recursive Conditioning Adnan Darwiche Computer Science Department UCLA](https://reader034.vdocument.in/reader034/viewer/2022052604/568143cf550346895db05cdc/html5/thumbnails/49.jpg)
.8
0 1
.45
Cache FactorsCache Factors
.9 1 0
0
cf(T) : [0,1]
![Page 50: Recursive Conditioning Adnan Darwiche Computer Science Department UCLA](https://reader034.vdocument.in/reader034/viewer/2022052604/568143cf550346895db05cdc/html5/thumbnails/50.jpg)
RC
RC(T,e)if T is a leaf node, return Lookup(T,e)y := instantiation of context(T)If cacheT[y] <> nil, return cacheT[y]
p := 0for each instantiation c of cutset(T)-E do
p := p + RC(Tl,ec) RC(Tr,ec)
If cacheT[y]? then cacheT[y] := p
return p
![Page 51: Recursive Conditioning Adnan Darwiche Computer Science Department UCLA](https://reader034.vdocument.in/reader034/viewer/2022052604/568143cf550346895db05cdc/html5/thumbnails/51.jpg)
calls(T)=cutset#(Tp) [cf(Tp) context#(Tp) + (1-cf(Tp)) calls(Tp)]
When cf(T)=0 for every node T:
calls(T) = cutset#(Tp) calls(Tp)
When cf(T)=1 for every node T:
calls(T) = cutset#(Tp) context#(Tp) = cluster#(Tp)
exact count for discrete cache factors
![Page 52: Recursive Conditioning Adnan Darwiche Computer Science Department UCLA](https://reader034.vdocument.in/reader034/viewer/2022052604/568143cf550346895db05cdc/html5/thumbnails/52.jpg)
![Page 53: Recursive Conditioning Adnan Darwiche Computer Science Department UCLA](https://reader034.vdocument.in/reader034/viewer/2022052604/568143cf550346895db05cdc/html5/thumbnails/53.jpg)
![Page 54: Recursive Conditioning Adnan Darwiche Computer Science Department UCLA](https://reader034.vdocument.in/reader034/viewer/2022052604/568143cf550346895db05cdc/html5/thumbnails/54.jpg)
![Page 55: Recursive Conditioning Adnan Darwiche Computer Science Department UCLA](https://reader034.vdocument.in/reader034/viewer/2022052604/568143cf550346895db05cdc/html5/thumbnails/55.jpg)
Extension to Dgraphs
)(A )|( BD)|( BC)|( AB
Marginals
B
![Page 56: Recursive Conditioning Adnan Darwiche Computer Science Department UCLA](https://reader034.vdocument.in/reader034/viewer/2022052604/568143cf550346895db05cdc/html5/thumbnails/56.jpg)
Dgraph
)(A )|( BD)|( BC)|( AB
AB
![Page 57: Recursive Conditioning Adnan Darwiche Computer Science Department UCLA](https://reader034.vdocument.in/reader034/viewer/2022052604/568143cf550346895db05cdc/html5/thumbnails/57.jpg)
DgraphAB A
)(A )|( BD)|( BC)|( AB
![Page 58: Recursive Conditioning Adnan Darwiche Computer Science Department UCLA](https://reader034.vdocument.in/reader034/viewer/2022052604/568143cf550346895db05cdc/html5/thumbnails/58.jpg)
Dgraph
)(A )|( BD)|( BC)|( AB
AB A BC
![Page 59: Recursive Conditioning Adnan Darwiche Computer Science Department UCLA](https://reader034.vdocument.in/reader034/viewer/2022052604/568143cf550346895db05cdc/html5/thumbnails/59.jpg)
Dgraph
)(A )|( BD)|( BC)|( AB
AB A BC BD
![Page 60: Recursive Conditioning Adnan Darwiche Computer Science Department UCLA](https://reader034.vdocument.in/reader034/viewer/2022052604/568143cf550346895db05cdc/html5/thumbnails/60.jpg)
Number of Dtree Internal Nodes
Network Dtree Dgraph
Water 31 122
Alarm 36 142
B 17 66
Barley 47 186
![Page 61: Recursive Conditioning Adnan Darwiche Computer Science Department UCLA](https://reader034.vdocument.in/reader034/viewer/2022052604/568143cf550346895db05cdc/html5/thumbnails/61.jpg)
SamIam Screenshot
![Page 62: Recursive Conditioning Adnan Darwiche Computer Science Department UCLA](https://reader034.vdocument.in/reader034/viewer/2022052604/568143cf550346895db05cdc/html5/thumbnails/62.jpg)
Discrete Cache Factor Search
• Problem– Given X MB of cache memory for RC– Find a discrete caching scheme which has a
minimal running time where the amount of memory used by all the dtree node caches is no larger than X
![Page 63: Recursive Conditioning Adnan Darwiche Computer Science Department UCLA](https://reader034.vdocument.in/reader034/viewer/2022052604/568143cf550346895db05cdc/html5/thumbnails/63.jpg)
Number of RC Calls
• Can estimate running time of RC• Function of:
– dtree– cache factors
pt
ppppp tcallstcftcontexttcftcutsettcalls *1**)(##
![Page 64: Recursive Conditioning Adnan Darwiche Computer Science Department UCLA](https://reader034.vdocument.in/reader034/viewer/2022052604/568143cf550346895db05cdc/html5/thumbnails/64.jpg)
Different Cache Factors
1
2 3
4 5 6 7
cf=0
cf=1cf=01
2 3
4 5 6 7
cf=0
cf=0cf=0
1
2 3
4 5 6 7
cf=1
cf=0cf=01
2 3
4 5 6 7
cf=0
cf=1cf=1
1
2 3
4 5 6 7
cf=0
cf=0cf=1
1
2 3
4 5 6 7
cf=1
cf=1cf=0
1
2 3
4 5 6 7
cf=1
cf=0cf=11
2 3
4 5 6 7
cf=1
cf=1cf=1
20 sec65 sec35 sec
75 sec80 sec90 sec
30 sec 10 sec
![Page 65: Recursive Conditioning Adnan Darwiche Computer Science Department UCLA](https://reader034.vdocument.in/reader034/viewer/2022052604/568143cf550346895db05cdc/html5/thumbnails/65.jpg)
Search Tree
1
2 3
4 5 6 7
cf=?
cf=?cf=?
1
2 3
4 5 6 7
cf=0
cf=?cf=?
No Caching
……
1
2 3
4 5 6 7
cf=1
cf=?cf=?
Full Caching
……
![Page 66: Recursive Conditioning Adnan Darwiche Computer Science Department UCLA](https://reader034.vdocument.in/reader034/viewer/2022052604/568143cf550346895db05cdc/html5/thumbnails/66.jpg)
Cost Function
1
2 3
4 5 6 7
cf=?
cf=?cf=?
1
2 3
4 5 6 7
cf=0
cf=?cf=?1
2 3
4 5 6 7
cf=1
cf=?cf=?
… …… …
![Page 67: Recursive Conditioning Adnan Darwiche Computer Science Department UCLA](https://reader034.vdocument.in/reader034/viewer/2022052604/568143cf550346895db05cdc/html5/thumbnails/67.jpg)
Cost Function
1
2 3
4 5 6 7
cf=?
cf=?cf=?
1
2 3
4 5 6 7
cf=0
cf=?cf=?1
2 3
4 5 6 7
cf=1
cf=1cf=1
… …… …
Admissible
![Page 68: Recursive Conditioning Adnan Darwiche Computer Science Department UCLA](https://reader034.vdocument.in/reader034/viewer/2022052604/568143cf550346895db05cdc/html5/thumbnails/68.jpg)
Search Tree Pruning
• Memory-based pruning– Prune any branch which assigns more than the
total memory assigned to the algorithm• Cost-based pruning
– Prune any node whose cost function f(n) is already larger than the current best goal node
![Page 69: Recursive Conditioning Adnan Darwiche Computer Science Department UCLA](https://reader034.vdocument.in/reader034/viewer/2022052604/568143cf550346895db05cdc/html5/thumbnails/69.jpg)
3.3 MB = 28 Sec7.6 MB = 12 Sec
![Page 70: Recursive Conditioning Adnan Darwiche Computer Science Department UCLA](https://reader034.vdocument.in/reader034/viewer/2022052604/568143cf550346895db05cdc/html5/thumbnails/70.jpg)
.56 MB = 59 Sec
2.6 MB = 5.9 Sec
1 MB = 15.7 Sec
.06 MB = 120 Sec
![Page 71: Recursive Conditioning Adnan Darwiche Computer Science Department UCLA](https://reader034.vdocument.in/reader034/viewer/2022052604/568143cf550346895db05cdc/html5/thumbnails/71.jpg)
.19 KB = 2 Sec2.8 KB < .01 Sec.23 KB = .7 Sec
![Page 72: Recursive Conditioning Adnan Darwiche Computer Science Department UCLA](https://reader034.vdocument.in/reader034/viewer/2022052604/568143cf550346895db05cdc/html5/thumbnails/72.jpg)
22 MB = 1 Min
6.5 MB = 4 Min
![Page 73: Recursive Conditioning Adnan Darwiche Computer Science Department UCLA](https://reader034.vdocument.in/reader034/viewer/2022052604/568143cf550346895db05cdc/html5/thumbnails/73.jpg)
370 MB = 22 Min
150 MB = 5.8 Hours
![Page 74: Recursive Conditioning Adnan Darwiche Computer Science Department UCLA](https://reader034.vdocument.in/reader034/viewer/2022052604/568143cf550346895db05cdc/html5/thumbnails/74.jpg)
400 MB = 18 Min
75 MB = 2 Hours
![Page 75: Recursive Conditioning Adnan Darwiche Computer Science Department UCLA](https://reader034.vdocument.in/reader034/viewer/2022052604/568143cf550346895db05cdc/html5/thumbnails/75.jpg)
Conclusion: Inference
• Alternative conditioning paradigm– condition to decompose– dtrees: decomposition policy
• New complexity results:– linear space: O(n cwlog n)– any-space: O(n) ---- O(n cw) O(n cwlog n) ---- O(n cw)