Beruflich Dokumente
Kultur Dokumente
Brian Booth
Every column
of D is a
prototype
Every column
of D is a
prototype
Sparse Coding in V1
Introduction The Basics Adding Prior Knowledge Conclusions
Outline
Outline
Outline
4 Conclusions
Introduction The Basics Adding Prior Knowledge Conclusions
Outline
4 Conclusions
Outline
4 Conclusions
Introduction The Basics Adding Prior Knowledge Conclusions
Learning:
Given training data xj , j {1, , m}
Learn dictionary D and sparse code
Introduction The Basics Adding Prior Knowledge Conclusions
Learning:
Given training data xj , j {1, , m}
Learn dictionary D and sparse code
Encoding:
Given test data x, dictionary D
Learn sparse code
2
subject to di c, i = 1, , n.
Introduction The Basics Adding Prior Knowledge Conclusions
2
subject to di c, i = 1, , n.
In matrix notation:
D,A
arg minX AD2F + |i,j |
i,j
subject to D2i,j c, i = 1, , n.
i
2
subject to di c, i = 1, , n.
In matrix notation:
D,A
arg minX AD2F + |i,j |
i,j
subject to D2i,j c, i = 1, , n.
i
D
arg minX AD2F
subject to D2i,j c, i = 1, , n.
i
D
arg minX AD2F
subject to D2i,j c, i = 1, , n.
i
j=1 i
Introduction The Basics Adding Prior Knowledge Conclusions
D
arg minX AD2F
subject to D2i,j c, i = 1, , n.
i
j=1 i
j=1 i
Introduction The Basics Adding Prior Knowledge Conclusions
j=1 i
D
D () = min L (D, ) =
j=1 i
T T T T
XA
D
D () = min L (D, ) = tr X X XA AA + c
1 T
Introduction The Basics Adding Prior Knowledge Conclusions
j=1 i
T T T T
XA
D
D () = min L (D, ) = tr X X XA AA + c
1 T
j=1 i
T T T T
XA
D
D () = min L (D, ) = tr X X XA AA + c
1 T
DT = AAT + XAT
1 T
T T T
D = AA + XA
1 T
A
arg minX AD2F + |i,j |
i,j
A
arg minX AD2F + |i,j |
i,j
A
arg minX AD2F + |i,j |
i,j
A
arg minX AD2F + |i,j |
i,j
Note:
Same problem as the encoding problem.
Runtime of optimization in the encoding stage?
Introduction The Basics Adding Prior Knowledge Conclusions
Outline
4 Conclusions
Introduction The Basics Adding Prior Knowledge Conclusions
Outline
4 Conclusions
Dictionaries are
over-complete bases
Introduction The Basics Adding Prior Knowledge Conclusions
Dictionaries are
over-complete bases
Dictate relationships
between atoms
Dictionaries are
over-complete bases
Dictate relationships
between atoms
Example: Hierarchical
dictionaries
Introduction The Basics Adding Prior Knowledge Conclusions
Problem Statement
Goal:
Have sub-groups of sparse code
all be non-zero (or zero).
Problem Statement
Goal:
Have sub-groups of sparse code
all be non-zero (or zero).
Hierarchical:
If a node is non-zero, its parent
must be non-zero
If a nodes parent is zero, the
node must be zero
Introduction The Basics Adding Prior Knowledge Conclusions
Problem Statement
Goal:
Have sub-groups of sparse code
all be non-zero (or zero).
Hierarchical:
If a node is non-zero, its parent
must be non-zero
If a nodes parent is zero, the
node must be zero
Implementation:
Change the regularization
Enforce sparsity differently...
Group Regularization
j j 2
arg min x D
j=1
D,{j }
m
Introduction The Basics Adding Prior Knowledge Conclusions
Group Regularization
j j 2
arg min x D + j
j=1
D,{j }
m
Group Regularization
j j 2
arg min x D + j
j=1
D,{j }
m
where
() = wg |g
gP
Introduction The Basics Adding Prior Knowledge Conclusions
Group Regularization
j j 2
arg min x D + j
j=1
D,{j }
m
where
() = wg |g
gP
Group Regularization
j j 2
arg min x D + j
j=1
D,{j }
m
where
() = wg |g
gP
Group Regularization
j j 2
arg min x D + j
j=1
D,{j }
m
where
() = wg |g
gP
Other Examples
Outline
4 Conclusions
Outline
4 Conclusions
Introduction The Basics Adding Prior Knowledge Conclusions
Summary
Summary
Two interesting
directions:
Introduction The Basics Adding Prior Knowledge Conclusions
Summary
Two interesting
directions:
Increasing
speed of the
testing phase
Summary
Two interesting
directions:
Increasing
speed of the
testing phase
Optimizing
dictionary
structure