2 CraigA.Knoblock UniversityofSouthernCalifornia 3 - - PowerPoint PPT Presentation

Craig
A.
Knoblock
 University
of
Southern
California


2


Craig
A.
Knoblock
 University
of
Southern
California


3


Craig
A.
Knoblock
 University
of
Southern
California


4


Craig
A.
Knoblock
 University
of
Southern
California


5


6


Building
Identification
(BID)
Problem


Traditional
Sources
 Non‐traditional
Sources
 Before
 After


Craig
A.
Knoblock
 University
of
Southern
California


7


[Michalowski & Knoblock 2005]

Craig
A.
Knoblock
 University
of
Southern
California


9


• Set of street names
• Set of buildings
• Potential street(s) it is on
• Side of street it is on
• Order for a given street
• Additional information
• Side of street where even

numbers lie

• Ascending addresses direction

10


Craig
A.
Knoblock
 University
of
Southern
California


11


Craig
A.
Knoblock
 University
of
Southern
California


12


Enforces all these buildings will be even or

• dd, not a mix

Craig
A.
Knoblock
 University
of
Southern
California


13


Enforces address > address because we know numbers ascend in south direction on N/S running streets

Craig
A.
Knoblock
 University
of
Southern
California


14


Street A

Enforces that all of the odd #s and the even #s for Street A are in the solution returned

Craig
A.
Knoblock
 University
of
Southern
California


15


16


El
Segundo
CA
 San
Francisco
CA
 Downtown
Los
Angeles
 New
Orleans
LA
 Belgrade
Serbia


17


Block
Numbering


YES
 NO


18


‘Addresses
increase
West’

CONFLICTS

‘Addresses
increase
East’
 Addresses
increase
West
 Addresses
increase
East


?


1
 2


Constraints
have
different
scopes


19


Generic
 Model
 Accurate
 Model
 Solutions


20


Problem
Instance


Input
information
 F =
{F1,F2,…,Fn}
 Generic
model
 CB =
{
C1,C2,…,Ci}


Refined
model:
Cnew
=
CB

∪
CI

Inference
Engine


Inference
rules






























 Rt =
{R1,R2,…,Rz}
 Rk:

Fi
∈
F

→
CI ∈
CL

Constraint
Library


User‐defined
(&
learned)
constraints
 CL =
{Cl1,Cl2,…,Clz}


21


22


Corner
building
can
only
be
on
one
street
 A
single
address
per
building
 1 3 5 7

23


BID
Problem
Sample
Library


• Odd
on
North
(or
South)

• Odd
on
East
(or
West)

• Ascending
North
(or
South)

• Ascending
East
(or
West)

• Block
Numbering

• Continuous
Numbering

• …


24


Constraint
Library


Applicable


Negative
 Positive
 Null
 Unknown


Non‐applicable


Status:
 Support:


25


Spatial
Separation
 Problem
Space


Support
Vector
Machines

[Vapnik,
1995]


26


Inferred
Model
 
 Constraint
1
 
 Constraint
2
 
 Constraint
3…
 Conflict
 Data
Points
 {D1,2,3,4,5,6,7,8,9,10}
 D1,2,5,6

Constraint
1
 D3,4,7,8

Constraint
2
 D9,10




?


Domain‐independent
solution


D2
 D5
 D1
 D6
 D3
 D8
 D7
 D4


Support
Vector
Machine
Model
 Constraint
1
 Constraint
2
 Class
Labels:


D9
 D10


Classify
 unknown
data
 points


Craig
A.
Knoblock
 University
of
Southern
California


27


Phone
Book
Linked 
 
to
Streets 


Constraint
Reasoning
to 
 Link
Data
to
Buildings 


Exploit
Maps
for
Disambigua>on 


Results
Propagated
to
 
 Further
Reduce
Ambiguity 


Craig
A.
Knoblock
 University
of
Southern
California


33


34


35


36