2 CraigA.Knoblock UniversityofSouthernCalifornia 3 - PowerPoint PPT Presentation

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
 Craig
A.
Knoblock
 University
of
Southern
California


Building
Identification
(BID)
Problem
 Traditional
Sources
 Non‐traditional
Sources
 Before
 After
 6


7
 Craig
A.
Knoblock
 University
of
Southern
California


[Michalowski & Knoblock 2005]

• 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 9
 Craig
A.
Knoblock
 University
of
Southern
California


10


11
 Craig
A.
Knoblock
 University
of
Southern
California


Enforces all these buildings will be even or odd, not a mix 12
 Craig
A.
Knoblock
 University
of
Southern
California


Enforces address > address because we know numbers ascend in south direction on N/S running streets 13
 Craig
A.
Knoblock
 University
of
Southern
California


Street A Enforces that all of the odd #s and the even #s for Street A are in the solution returned 14
 Craig
A.
Knoblock
 University
of
Southern
California


15
 Craig
A.
Knoblock
 University
of
Southern
California


El
Segundo
CA
 Downtown
Los
Angeles
 Belgrade
Serbia
 16
 San
Francisco
CA
 New
Orleans
LA


Block
Numbering
 YES
 NO
 17


?
 1
 2
 Constraints
have
different
scopes
 ‘Addresses
increase 
 West’

CONFLICTS

‘Addresses
increase
East’
 Addresses
increase
West
 Addresses
increase
East
 18


Generic
 Model 
 Accurate
 Model 
 Solutions 
 19


Constraint
Library
 Problem
Instance
 User‐defined
(&
learned)
constraints
 Input
information
 Generic
model
 C L =
{C l1 ,C l2 ,…,C lz }
 F =
{F 1 ,F 2 ,…,F n }
 C B =
{
C 1 ,C 2 ,…,C i }
 Inference
Engine
 Inference
rules






























 R t =
{R 1 ,R 2 ,…,R z } 
 R k :

 F i 
 ∈ 
 F 

 → 
 C I ∈ 
 C L Refined
model : 
 C new 
=
 C B 

 ∪ 
 C I 20


21


3 5 1 7 Corner
building
can
only
be
on
one
street
 A
single
address
per
building
 22


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
 • …
 23


Constraint
Library
 Status:
 Unknown
 Applicable
 Non‐applicable
 Support:
 Positive
 Null
 Negative
 24


Spatial
Separation
 Problem
Space
 25
 Support
Vector
Machines

 [Vapnik,
1995]


Domain‐independent
solution
 Constraint
1
 Class
Labels:
 Inferred
Model
 Constraint
2
 
 Constraint
1
 Conflict
 
 Constraint
2
 D2
 D7
 
 Constraint
3…
 Classify
 D10
 D1
 unknown
data
 D5
 D3
 points
 Data
Points
 D6
 D8
 {D 1,2,3,4,5,6,7,8,9,10 }
 D4
 D 1,2,5,6 
  
Constraint
1
 D9
 D 3,4,7,8 
  
Constraint
2
 D 9,10 



  
?
 26
 Support
Vector
Machine
Model


27
 Craig
A.
Knoblock
 University
of
Southern
California


Phone
Book
Linked 
 
to
Streets 


Constraint
Reasoning
to 
 Link
Data
to
Buildings 


Exploit
Maps
for
Disambigua>on 


Results
Propagated
to
 
 Further
Reduce
Ambiguity 


33
 Craig
A.
Knoblock
 University
of
Southern
California


34


35


36