Ce diaporama a bien été signalé.
Nous utilisons votre profil LinkedIn et vos données d’activité pour vous proposer des publicités personnalisées et pertinentes. Vous pouvez changer vos préférences de publicités à tout moment.
Mathématiques - Informatique
Optimal Transport
From a Computer Perspective
Summer School « Geometric Measure Theory and
Ca...
Part. 1. Goals and Motivations - Understanding
Part. 2. Optimal Transport Elementary intro.
Part. 3. The Semi-Discrete Cas...
OVERVIEW
Cédric Villani
Optimal Transport Old & New
Topics on Optimal Transport
Yann Brenier
The polar factorization theor...
OVERVIEW
June 2015
Institut Fourier
March 2015
Bonn
Febr 2015
BIRS (Canada)
Febr 2015
LJLL
Goals and Motivations
1
Part. 1 Optimal Transport
Measuring distances between functions
Part. 1 Optimal Transport
Measuring distances between functions
Part. 1 Optimal Transport
Measuring distances between functions
Part. 1 Optimal Transport
Measuring distances between function
Part. 1 Optimal Transport
Interpolating functions
Part. 1 Optimal Transport
Interpolating functions
Part. 1 Optimal Transport
Interpolating functions
Part. 1 Optimal Transport
Interpolating functions
Part. 1 Optimal Transport
Interpolating functions
Part. 1 Optimal Transport
Interpolating functions
Part. 1 Optimal Transport
Interpolating functions
Part. 1 Optimal Transport
Interpolating functions
Part. 1 Optimal Transport
Interpolating functions
Part. 1 Optimal Transport
Interpolating functions
Part. 1 Optimal Transport
Interpolating functions
Part. 1 Optimal Transport
Interpolating functions
Part. 1 Optimal Transport
?
T
Part. 1 Optimal Transport
?
T
distance between the shapes
Part. 1 Optimal Transport
while preserving mass and minimizing the effort ?
?
T
Part. 1 Optimal Transport
while preserving mass and minimizing the effort ?
?
T
Part. 1 Optimal Transport
while preserving mass and minimizing the effort ?
?
T
Part. 1 Optimal Transport
OT=
Yann Brenier:
it can be replaced with the Monge-Ampère
Part. 1 Optimal Transport
OT=
Yann Brenier:
it can be replaced with the Monge-Ampère
New ways of simulating physics with a...
Part. 1 Optimal Transport
OT=
Yann Brenier:
Laplace operator is used in a PDE,
it can be replaced with the Monge-Ampère
Ne...
Part. 1 Optimal Transport
OT=
Yann Brenier:
Laplace operator is used in a PDE,
it can be replaced with the Monge-Ampère
Ne...
Part. 1 Optimal Transport
Gaspard Monge - 1784
Part. 1 Optimal Transport
Gaspard Monge geometry and light
ANR TOMMI Workshop
Part. 1 Optimal Transport
[Castro, Merigot, Thibert 2014]
Optimal transport
geometry and light
[Caffarelli, Kochengin, and...
Part. 1 Optimal Transport Image Processing
Barycenters / mixing textures Video-style transfer
[Nicolas Bonneel, Julien Rab...
Part. 1 Optimal Transport
Optimal transport - geometry and light
[Chwartzburg, Testuz, Tagliasacchi, Pauly, SIGGRAPH 2014]
Part. 1. Motivations
Discretization of functionals involving the Monge-Ampère operator,
Benamou, Carlier, Mérigot, Oudet
a...
Part. 1 Optimal Transport
Optimal transport E.U.R.
The data
Part. 1 Optimal Transport
The millenium simulation project, Max Planck Institute fur Astrophysik
pc/h : parsec (= 3.2 ligh...
Part. 1 Optimal Transport
Optimal transport E.U.R.
The universal swimming pool
Part. 1 Optimal Transport
Optimal transport E.U.R.
let there be light !
The millenium simulation project,
Max Planck Insti...
Part. 1 Optimal Transport
Optimal transport E.U.R.
let there be light !
The millenium simulation project,
Max Planck Insti...
Part. 1 Optimal Transport
Optimal transport E.U.R.
let there be light !
The millenium simulation project,
Max Planck Insti...
Optimal Transport
an elementary introduction
2
Part. 2 Optimal Transport problem
(X; ) (Y; )
Two measures , such that d (x) = d (x)
X Y
Part. 2 Optimal Transport problem
A map T is a transport map between and if
(T-1(B)) = (B) for any Borel subset B of Y
(X;...
Part. 2 Optimal Transport problem
A map T is a transport map between and if
(T-1(B)) = (B) for any Borel subset B
B
(X; ) ...
Part. 2 Optimal Transport problem
A map T is a transport map between and if
(T-1(B)) = (B) for any Borel subset B
B
T-1(B)...
Part. 2 Optimal Transport problem
A map T is a transport map between and if
(T-1(B)) = (B) for any Borel subset B
(or = T#...
Part. 2 Optimal Transport problem
problem:
Find a transport map T that minimizes C(T) = X || x T(x) ||2 d (x)
(X; ) (Y; )
Part. 2 Optimal Transport problem
problem:
Find a transport map T that minimizes C(T) = X || x T(x) ||2 d (x)
Difficult to...
Part. 2 Optimal Transport problem
problem:
Find a transport map T that minimizes C(T) = X || x T(x) ||2 d (x)
Transport fr...
Part. 2 Optimal Transport problem
problem:
Find a transport map T that minimizes C(T) = X || x T(x) ||2 d (x)
Transport fr...
Part. 2 Optimal Transport problem
problem:
Find a transport map T that minimizes C(T) = X || x T(x) ||2 d (x)
Transport fr...
Part. 2 Optimal Transport problem
problem:
Find a transport map T that minimizes C(T) = X || x T(x) ||2 d (x)
Transport fr...
Part. 2 Optimal Transport Kantorovich
problem:
Find a transport map T that minimizes C(T) = X || x T(x) ||2 d (x)
(1942):
...
Part. 2 Optimal Transport Kantorovich
problem:
Find a transport map T that minimizes C(T) = X || x T(x) ||2 d (x)
Find a m...
Part. 2 Optimal Transport Kantorovich
problem:
Find a transport map T that minimizes C(T) = X || x T(x) ||2 d (x)
Find a m...
Part. 2 Optimal Transport Kantorovich
problem:
Find a transport map T that minimizes C(T) = X || x T(x) ||2 d (x)
Find a m...
Part. 2 Optimal Transport Kantorovich
Transport plan example 1/4 : translation of a segment
Part. 2 Optimal Transport Kantorovich
Transport plan example 1/4 : translation of a segment
Part. 2 Optimal Transport Kantorovich
Transport plan example 2/4 : spitting a segment
Part. 2 Optimal Transport Kantorovich
Part. 2 Optimal Transport Kantorovich
Part. 2 Optimal Transport Kantorovich
Part. 2 Optimal Transport Kantorovich
Transport plan example 3/4 : splitting a Dirac into two Diracs
Part. 2 Optimal Transport Kantorovich
Transport plan example 3/4 : splitting a Dirac into two Diracs
(No transport map)
Part. 2 Optimal Transport Kantorovich
Transport plan example 4/4 : splitting a Dirac into two segments
Part. 2 Optimal Transport Kantorovich
Transport plan example 4/4 : splitting a Dirac into two segments
(No transport map)
Part. 2 Optimal Transport Duality
Part. 2 Optimal Transport Duality
Duality is easier to understand with a discrete version
Part. 2 Optimal Transport Duality
(DMK):
Min <c, >
P1 = u
s.t. P2 = v
Part. 2 Optimal Transport Duality
(DMK):
Min <c, >
P1 = u
s.t. P2 = v
=
11
12
1n
22
2n
mn
IRmn
Part. 2 Optimal Transport Duality
(DMK):
Min <c, >
P1 = u
s.t. P2 = v
=
11
12
1n
22
2n
mn
c =
c11
c12
c1n
c22
c2n
cmn
IRmn...
Part. 2 Optimal Transport Duality
(DMK):
Min <c, >
P1 = u
s.t. P2 = v
=
11
12
1n
22
2n
mn
c =
c11
c12
c1n
c22
c2n
cmn
cij ...
Part. 2 Optimal Transport Duality
(DMK):
Min <c, >
P1 = u
s.t. P2 = v
0
=
11
12
1n
22
2n
mn
c =
c11
c12
c1n
c22
c2n
cmn
ci...
Part. 2 Optimal Transport Duality
(DMK):
Min <c, >
P1 = u
s.t. P2 = v
=
11
12
1n
22
2n
mn
c =
c11
c12
c1n
c22
c2n
cmn
cij ...
Part. 2 Optimal Transport Duality
(DMK):
Min <c, >
P1 = u
s.t. P2 = v
=
11
12
1n
22
2n
mn
c =
c11
c12
c1n
c22
c2n
cmn
IRmn...
Part. 2 Optimal Transport Duality
Consider L( , ) = <c, > - < , P1 - u> - < , P2 - v>
(DMK):
Min <c, >
P1 = u
s.t. P2 = v<...
Part. 2 Optimal Transport Duality
Consider L( , ) = <c, > - < , P1 - u> - < , P2 - v>
Remark: Sup[ L( , ) ] = < c, > if P1...
Part. 2 Optimal Transport Duality
Consider L( , ) = <c, > - < , P1 - u> - < , P2 - v>
Remark: Sup[ L( , ) ] = < c, > if P1...
Part. 2 Optimal Transport Duality
Consider L( , ) = <c, > - < , P1 - u> - < , P2 - v>
Remark: Sup[ L( , ) ] = < c, > if P1...
Part. 2 Optimal Transport Duality
Consider L( , ) = <c, > - < , P1 - u> - < , P2 - v>
Remark: Sup[ L( , ) ] = < c, > if P1...
Part. 2 Optimal Transport Duality
Consider L( , ) = <c, > - < , P1 - u> - < , P2 - v>
Remark: Sup[ L( , ) ] = < c, > if P1...
Part. 2 Optimal Transport Duality
IRm
IRn
Inf [ Sup[ <c, > - < , P1 - u> - < , P2 - v> ] ]
(DMK):
Min <c, >
P1 = u
s.t. P2...
Part. 2 Optimal Transport Duality
IRm
IRn
0
Inf [ Sup[ <c, > - < , P1 - u> - < , P2 - v> ] ]
IRm
IRn
Sup[ Inf[ <c, > - < ,...
Part. 2 Optimal Transport Duality
IRm
IRn
Inf [ Sup[ <c, > - < , P1 - u> - < , P2 - v> ] ]
IRm
IRn
Sup[ Inf[ <c, > - < , P...
Part. 2 Optimal Transport Duality
IRm
IRn
Inf [ Sup[ <c, > - < , P1 - u> - < , P2 - v> ] ]
IRm
IRn
Sup[ Inf[ <c, > - < , P...
Part. 2 Optimal Transport Duality
IRm
IRn
Sup[ Inf[ < ,c-P1
t P2
t > + < ,u> + < , v> ] ]
Interpret
Sup[ < ,u> + < , v> ]
...
Part. 2 Optimal Transport Duality
IRm
IRn
Sup[ Inf[ < ,c-P1
t P2
t > + < ,u> + < , v> ] ]
Interpret
Sup[ < ,u> + < , v> ]
...
Part. 2 Optimal Transport Kantorovich dual
Find a measure defined on X x Y
such that x in X d (x,y) = d (x)
and y in Y d (...
Part. 2 Optimal Transport Kantorovich dual
Find a measure defined on X x Y
such that x in X d (x,y) = d (x)
and y in Y d (...
Part. 2 Optimal Transport Kantorovich dual
Find a measure defined on X x Y
such that x in X d (x,y) = d (x)
and y in Y d (...
Part. 2 Optimal Transport Kantorovich dual
Find a measure defined on X x Y
such that x in X d (x,y) = d (x)
and y in Y d (...
Part. 2 Optimal Transport Kantorovich dual
Find a measure defined on X x Y
such that x in X d (x,y) = d (x)
and y in Y d (...
Part. 2 Optimal Transport Kantorovich dual
Find a measure defined on X x Y
such that x in X d (x,y) = d (x)
and y in Y d (...
Part. 2 Optimal Transport c-conjugate functions
Find two functions in L1( ) and in L1( )
Such that for all x,y, (x) + y ||...
Part. 2 Optimal Transport c-conjugate functions
Find two functions in L1( ) and in L1( )
Such that for all x,y, (x) + y ||...
Part. 2 Optimal Transport c-conjugate functions
Find two functions in L1( ) and in L1( )
Such that for all x,y, (x) + y ||...
Part. 2 Optimal Transport c-conjugate functions
Find a c-concave function
that maximizes X (x)d + Y
c(y)d
Part. 2 Optimal Transport c-conjugate functions
Find a c-concave function
that maximizes X (x)d + Y
c(y)d
is called a
Part. 2 Optimal Transport c-subdifferential
What about our initial problem ?
is called a
Find a c-concave function
that ma...
Part. 2 Optimal Transport c-subdifferential
is called a
Find a c-concave function
that maximizes X (x)d + Y
c(y)d
Part. 2 Optimal Transport c-subdifferential
Part. 2 Optimal Transport c-subdifferential
Proof: see OTON, chap. 10.
Part. 2 Optimal Transport c-subdifferential
Proof: see OTON, chap. 10.
Heuristic argument (at the beginning of the same ch...
Part. 2 Optimal Transport c-subdifferential
Proof: see OTON, chap. 10.
Heuristic argument (at the beginning of the same ch...
Part. 2 Optimal Transport c-subdifferential
Proof: see OTON, chap. 10.
Heuristic argument (at the beginning of the same ch...
Part. 2 Optimal Transport c-subdifferential
Proof: see OTON, chap. 10.
Heuristic argument (at the beginning of the same ch...
Part. 2 Optimal Transport c-subdifferential
Proof: see OTON, chap. 10.
Heuristic argument (at the beginning of the same ch...
Part. 2 Optimal Transport c-subdifferential
Proof: see OTON, chap. 10.
Heuristic argument (at the beginning of the same ch...
Part. 2 Optimal Transport c-subdifferential
Proof: see OTON, chap. 10.
Heuristic argument (at the beginning of the same ch...
Part. 2 Optimal Transport c-subdifferential
Proof: see OTON, chap. 10.
Heuristic argument (at the beginning of the same ch...
Part. 2 Optimal Transport c-subdifferential
Find a c-concave function
that maximizes X (x)d + Y
c(y)d
Part. 2 Optimal Transport c-subdifferential
Find a c-concave function
that maximizes X (x)d + Y
c(y)d
{
grad (x) with (x) ...
Part. 2 Optimal Transport convexity
Find a c-concave function
that maximizes X (x)d + Y
c(y)d
{
grad (x) with (x) := (½ x2...
Part. 2 Optimal Transport convexity
Find a c-concave function
that maximizes X (x)d + Y
c(y)d
{
grad (x) with (x) := (½ x2...
Part. 2 Optimal Transport convexity
Find a c-concave function
that maximizes X (x)d + Y
c(y)d
{
grad (x) with (x) := (½ x2...
Part. 2 Optimal Transport no collision
If T(.) exists, then
T(x) = x grad (x) = grad (½ x2- (x) )
{grad (x)
Find a c-conca...
Part. 2 Optimal Transport no collision
If T(.) exists, then
T(x) = x grad (x) = grad (½ x2- (x) )
{grad (x)
Find a c-conca...
Part. 2 Optimal Transport no collision
If T(.) exists, then
T(x) = x grad (x) = grad (½ x2- (x) )
{grad (x)
Find a c-conca...
Part. 2 Optimal Transport Monge-Ampere
What about our initial problem ? If T(.) exists, then one can show that:
T(x) = x g...
Part. 2 Optimal Transport Monge-Ampere
What about our initial problem ? If T(.) exists, then one can show that:
T(x) = x g...
Part. 2 Optimal Transport Monge-Ampere
What about our initial problem ?
T(x) = x grad (x) = grad (½ x2- (x) )
{grad (x) wi...
Part. 2 Optimal Transport summary
Find a transport map T that minimizes C(T) = X || x T(x) ||2 d (x)
Part. 2 Optimal Transport summary
Find a transport map T that minimizes C(T) = X || x T(x) ||2 d (x)
(Kantorovich formulat...
Part. 2 Optimal Transport summary
Find a transport map T that minimizes C(T) = X || x T(x) ||2 d (x)
(Kantorovich formulat...
Part. 2 Optimal Transport summary
Find a transport map T that minimizes C(T) = X || x T(x) ||2 d (x)
(Kantorovich formulat...
Part. 2 Optimal Transport Isoperimetric inequality
For a given volume,
ball is the shape that minimizes border area
Part. 2 Optimal Transport Isoperimetric inequality
Vol(B2
n)1/n ( f n/(n-1))(n-1)/n
L1 Sobolev inegality: Given f: IRn suf...
Part. 2 Optimal Transport Isoperimetric inequality
n Vol(B2
n)1/n ( f n/(n-1))(n-1)/n
L1 Sobolev inegality: Given f: IRn s...
Part. 2 Optimal Transport Isoperimetric inequality
Vol(B2
n)1/n ( f n/(n-1))(n-1)/n
L1 Sobolev inegality: Given f: IRn suf...
Part. 2 Optimal Transport Isoperimetric inequality
Vol(B2
n)1/n ( f n/(n-1))(n-1)/n
L1 Sobolev inegality: Given f: IRn suf...
Part. 2 Optimal Transport Isoperimetric inequality
Vol(B2
n)1/n ( f n/(n-1))(n-1)/n
L1 Sobolev inegality: Given f: IRn suf...
Part. 2 Optimal Transport Isoperimetric inequality
Vol(B2
n)1/n ( f n/(n-1))(n-1)/n
L1 Sobolev inegality: Given f: IRn suf...
Part. 2 Optimal Transport Isoperimetric inequality
Vol(B2
n)1/n ( f n/(n-1))(n-1)/n
L1 Sobolev inegality: a proof with OT ...
Part. 2 Optimal Transport Isoperimetric inequality
Vol(B2
n)1/n ( f n/(n-1))(n-1)/n
L1 Sobolev inegality: a proof with OT ...
Part. 2 Optimal Transport Isoperimetric inequality
Vol(B2
n)1/n ( f n/(n-1))(n-1)/n
L1 Sobolev inegality: a proof with OT ...
Part. 2 Optimal Transport Isoperimetric inequality
Vol(B2
n)1/n ( f n/(n-1))(n-1)/n
L1 Sobolev inegality: a proof with OT ...
Part. 2 Optimal Transport Isoperimetric inequality
Vol(B2
n)1/n ( f n/(n-1))(n-1)/n
L1 Sobolev inegality: a proof with OT ...
Part. 2 Optimal Transport Isoperimetric inequality
Vol(B2
n)1/n ( f n/(n-1))(n-1)/n
L1 Sobolev inegality: a proof with OT ...
Part. 2 Optimal Transport Isoperimetric inequality
Vol(B2
n)1/n ( f n/(n-1))(n-1)/n
L1 Sobolev inegality: a proof with OT ...
Part. 2 Optimal Transport Isoperimetric inequality
Vol(B2
n)1/n ( f n/(n-1))(n-1)/n
L1 Sobolev inegality: a proof with OT ...
Part. 2 Optimal Transport Isoperimetric inequality
Vol(B2
n)1/n ( f n/(n-1))(n-1)/n
L1 Sobolev inegality: a proof with OT ...
Part. 2 Optimal Transport Isoperimetric inequality
Vol(B2
n)1/n ( f n/(n-1))(n-1)/n
L1 Sobolev inegality: a proof with OT ...
Part. 2 Optimal Transport Isoperimetric inequality
Vol(B2
n)1/n ( f n/(n-1))(n-1)/n
L1 Sobolev inegality: a proof with OT ...
Part. 2 Optimal Transport Isoperimetric inequality
Vol(B2
n)1/n ( f n/(n-1))(n-1)/n
L1 Sobolev inegality: a proof with OT ...
Part. 2 Optimal Transport Isoperimetric inequality
Vol(B2
n)1/n ( f n/(n-1))(n-1)/n
L1 Sobolev inegality: a proof with OT ...
Semi-Discrete Optimal Transport
3
Part. 3 Optimal Transport how to program ?
Continuous
(X; ) (Y; )
Part. 3 Optimal Transport how to program ?
Continuous
Semi-discrete
(X; ) (Y; )
Part. 3 Optimal Transport how to program ?
Continuous
Semi-discrete
Discrete
(X; ) (Y; )
Part. 3 Optimal Transport how to program ?
Continuous
Semi-discrete
Discrete
(X; ) (Y; )
Part. 3 Optimal Transport semi-discrete
X
c (x)d + Y (y)d
Sup
c
(DMK)
(X; ) (Y; )
Part. 3 Optimal Transport semi-discrete
(X; ) (Y; )
j (yj) vj
X
c (x)d + Y (y)d
Sup
c
(DMK)
Part. 3 Optimal Transport semi-discrete
X
c (x)d + Y (y)d
Sup
c
(DMK)
j (yj) vj
Part. 3 Optimal Transport semi-discrete
X
c (x)d + Y (y)d
Sup
c
(DMK)
X inf yj Y [ || x yj ||2 - (yj) ] d
j (yj) vj
Part. 3 Optimal Transport semi-discrete
X
c (x)d + Y (y)d
Sup
c
(DMK)
j Lag (yj) || x yj ||2 - (yj) d
X inf yj Y [ || x yj...
Part. 3 Optimal Transport semi-discrete
G( ) = j Lag (yj) || x yj ||2 - (yj) d + j (yj) vj
Sup
c
(DMK)
Where: Lag (yj) = {...
Part. 3 Optimal Transport semi-discrete
G( ) = j Lag (yj) || x yj ||2 - (yj) d + j (yj) vj
Sup
c
(DMK)
Where: Lag (yj) = {...
Part. 3 Optimal Transport semi-discrete
G( ) = j Lag (yj) || x yj ||2 - (yj) d + j (yj) vj
Sup
c
(DMK)
Where: Lag (yj) = {...
Part. 3 Optimal Transport semi-discrete
G( ) = j Lag (yj) || x yj ||2 - (yj) d + j (yj) vj
Sup
c
(DMK)
Where: Lag (yj) = {...
Part. 3 Optimal Transport semi-discrete
G( ) = j Lag (yj) || x yj ||2 - (yj) d + j (yj) vj
Sup
c
(DMK)
Where: Lag (yj) = {...
Voronoi diagram: Vor(xi) = { x | d2(x,xi) < d2(x,xj) }
Part. 3 Power Diagrams
Power diagram: Pow(xi) = { x | d2(x,xi) i < d2(x,xj) j }
Voronoi diagram: Vor(xi) = { x | d2(x,xi) < d2(x,xj) }
Part. 3 Po...
Part. 3 Power Diagrams
Part. 3 Optimal Transport
Theorem: (direct consequence of MK duality
alternative proof in [Aurenhammer, Hoffmann, Aronov 9...
Part. 3 Optimal Transport
Theorem: (direct consequence of MK duality
alternative proof in [Aurenhammer, Hoffmann, Aronov 9...
Part. 3 Optimal Transport
Theorem: (direct consequence of MK duality
alternative proof in [Aurenhammer, Hoffmann, Aronov 9...
Part. 3 Optimal Transport the AHA paper
Idea of the proof
Consider the function fT(W) = (|| x T(x) ||2 (T(X)))d (x)
The (u...
Part. 3 Optimal Transport the AHA paper
Idea of the proof
Consider the function fT(W) = (|| x T(x) ||2 (T(X)))d (x)
T : an...
Part. 3 Optimal Transport the AHA paper
Idea of the proof
Consider the function fT(W) = (|| x T(x) ||2 (T(X)))d (x)
T : an...
Part. 3 Optimal Transport the AHA paper
Idea of the proof
Consider the function fT(W) = (|| x T(x) ||2 (T(X)))d (x)
T : an...
Part. 3 Optimal Transport the AHA paper
Idea of the proof
Consider the function fT(W) = (|| x T(x) ||2 (T(X)))d (x)
T : an...
Part. 3 Optimal Transport the AHA paper
Idea of the proof
Consider the function fT(W) = (|| x T(x) ||2 (T(X)))d (x)
W
fT(W...
Part. 3 Optimal Transport the AHA paper
Idea of the proof
Consider the function fT(W) = (|| x T(x) ||2 (T(X)))d (x)
W
fT(W...
Part. 3 Optimal Transport the AHA paper
Idea of the proof
Consider the function fT(W) = (|| x T(x) ||2 (T(X)))d (x)
fT(W) ...
Part. 3 Optimal Transport the AHA paper
Idea of the proof
Consider the function fT(W) = (|| x T(x) ||2 (T(X)))d (x)
fT(W) ...
Part. 3 Optimal Transport the AHA paper
Idea of the proof
Consider the function fT(W) = (|| x T(x) ||2 (T(X)))d (x)
fT(W) ...
Part. 3 Optimal Transport the AHA paper
Idea of the proof
Consider the function fT(W) = (|| x T(x) ||2 (T(X)))d (x)
fT(W) ...
Part. 3 Optimal Transport the AHA paper
Idea of the proof
Consider the function fT(W) = (|| x T(x) ||2 (T(X)))d (x)
fT(W) ...
Part. 3 Optimal Transport the AHA paper
fT(W)
W
fT(W)
f TW
(W) = minT fT(W)
Consider g(W) = f TW
vj j
vj - pow(yj)|| x yj ...
Part. 3 Optimal Transport the algorithm
Semi-discrete OT Summary:
X
c (x)d + Y (y)d
Sup
c
(DMK) G( ) =
Part. 3 Optimal Transport the algorithm
Semi-discrete OT Summary:
G( ) = g(W) = j Lag (yj) || x yj ||2 - (yj) d + j (yj) v...
Part. 3 Optimal Transport the algorithm
Semi-discrete OT Summary:
G( ) = g(W) = j Lag (yj) || x yj ||2 - (yj) d + j (yj) v...
Part. 3 Optimal Transport the algorithm
Semi-discrete OT Summary:
G( ) = g(W) = j Lag (yj) || x yj ||2 - (yj) d + j (yj) v...
Part. 3 Optimal Transport the algorithm
The [AHA] paper summary:
The optimal weights minimize a convex function
The gradie...
Part. 3 Optimal Transport the algorithm
The [AHA] paper summary:
The optimal weights minimize a convex function
The gradie...
Part. 3 Optimal Transport the algorithm
The [AHA] paper summary:
The optimal weights minimize a convex function
The gradie...
Part. 3 Optimal Transport the algorithm
The [AHA] paper summary:
The optimal weights minimize a convex function
The gradie...
Part. 3 Optimal Transport the algorithm
The [AHA] paper summary:
The optimal weights minimize a convex function
The gradie...
Part. 3 Optimal Transport the algorithm
The [AHA] paper summary:
The optimal weights minimize a convex function
The gradie...
Part. 3 Optimal Transport the algorithm
Compute Pow( (wi, si) M1 [Nivoliers, L 2014, Curves and Surfaces]
Implementation i...
Part. 3 Optimal Transport the algorithm
The [AHA] paper summary:
The optimal weights minimize a convex function
The gradie...
Part. 3 Optimal Transport the algorithm
The [AHA] paper summary:
The optimal weights minimize a convex function
The gradie...
Part. 3 Optimal Transport the algorithm
The [AHA] paper summary:
The optimal weights minimize a convex function
The gradie...
Part. 3 Optimal Transport the algorithm
The [AHA] paper summary:
The optimal weights minimize a convex function
The gradie...
Part. 3 Optimal Transport the algorithm
The [AHA] paper summary:
The optimal weights minimize a convex function
The gradie...
Part. 3 Optimal Transport the algorithm
The [AHA] paper summary:
The optimal weights minimize a convex function
The gradie...
Part. 3 Optimal Transport the algorithm
The [AHA] paper summary:
The optimal weights minimize a convex function
The gradie...
Understanding Going
4
Part. 4 Optimal Transport ???
Wait a minute:
This means that one can move (and possibly deform)
a power diagram simply by ...
Part. 4 Optimal Transport ???
Wait a minute:
This means that one can move (and possibly deform)
a power diagram simply by ...
Part. 4 Power Diagrams & Transport
Part. 4 Power Diagrams & Transport
Part. 4 Power Diagrams & Transport
Part. 4 Power Diagrams & Transport
Part. 4 Power Diagrams & Transport
Part. 4 Power Diagrams & Transport
Part. 4 Power Diagrams & Transport
hi
Part. 4 Power Diagrams & Transport
Part. 4 Power Diagrams & Transport
Part. 4 Power Diagrams & Transport
Part. 4 Power Diagrams & Transport
Part. 4 Optimal Transport still lost ? 2D examples
Numerical Experiment: Simple translation
Part. 4 Power Diagrams & Transport
Part. 4 Power Diagrams and Transport
Part. 4 Optimal Transport 2D examples
Numerical Experiment: Splitting a disk
Part. 4 Optimal Transport 2D examples
Numerical Experiment: A disk becomes two disks
Part. 4 Optimal Transport 3D examples
Numerical Experiment: A sphere becomes a cube
Part. 4 Optimal Transport 3D examples
Numerical Experiment: A sphere becomes two spheres
Part. 4 Optimal Transport 3D examples
Numerical Experiment: Armadillo to sphere
Part. 4 Optimal Transport 3D examples
Numerical Experiment: Other examples
Part. 4 Optimal Transport 3D examples
Numerical Experiment: Varying density
Part. 4 Optimal Transport 3D examples
Numerical Experiment: Performances
Part. 4 Optimal Transport 3D examples
Numerical Experiment: Performances
Note that a few years ago, several hours of super...
Part. 4 Optimal Transport 3D examples
Numerical Experiment: Performances
Note that a few years ago, several hours of super...
Part. 4 Optimal Transport 3D examples
Numerical Experiment: Performances
Note that a few years ago, several hours of super...
Part. 4 Optimal Transport 3D examples
Numerical Experiment: Performances
Note that a few years ago, several hours of super...
Conclusions Open questions
* Connections with physics, Legendre transform and entropy ?
[Cuturi & Peyré] regularized discr...
Conclusions - References
Some references (that this presentation is based on)
A Multiscale Approach to Optimal Transport,
...
Conclusions - References
Other references (that I need to read and understand)
Polar factorization and monotone rearrangem...
Online resources
All the sourcecode/documentation available from:
alice.loria.fr/software/geogram
Computes semi-discrete O...
Acknowledgements
Funding: European Research Council french
GOODSHAPE ERC-StG-205693
VORPALINE ERC-PoC-334829
ANR MORPHO, A...
Prochain SlideShare
Chargement dans…5
×

Optimal Transport for a Computer Programmer's Point of View

1 447 vues

Publié le

Course on Optimal Transport Theory and its implementation on a computer using the Semi-Discrete formulation.

Publié dans : Sciences
  • DOWNLOAD FULL BOOKS, INTO AVAILABLE FORMAT ......................................................................................................................... ......................................................................................................................... 1.DOWNLOAD FULL. PDF EBOOK here { https://tinyurl.com/yxufevpm } ......................................................................................................................... 1.DOWNLOAD FULL. EPUB Ebook here { https://tinyurl.com/yxufevpm } ......................................................................................................................... 1.DOWNLOAD FULL. doc Ebook here { https://tinyurl.com/yxufevpm } ......................................................................................................................... 1.DOWNLOAD FULL. PDF EBOOK here { https://tinyurl.com/yxufevpm } ......................................................................................................................... 1.DOWNLOAD FULL. EPUB Ebook here { https://tinyurl.com/yxufevpm } ......................................................................................................................... 1.DOWNLOAD FULL. doc Ebook here { https://tinyurl.com/yxufevpm } ......................................................................................................................... ......................................................................................................................... ......................................................................................................................... .............. Browse by Genre Available eBooks ......................................................................................................................... Art, Biography, Business, Chick Lit, Children's, Christian, Classics, Comics, Contemporary, Cookbooks, Crime, Ebooks, Fantasy, Fiction, Graphic Novels, Historical Fiction, History, Horror, Humor And Comedy, Manga, Memoir, Music, Mystery, Non Fiction, Paranormal, Philosophy, Poetry, Psychology, Religion, Romance, Science, Science Fiction, Self Help, Suspense, Spirituality, Sports, Thriller, Travel, Young Adult,
       Répondre 
    Voulez-vous vraiment ?  Oui  Non
    Votre message apparaîtra ici
  • DOWNLOAD FULL eBOOK INTO AVAILABLE FORMAT ......................................................................................................................... ......................................................................................................................... 1.DOWNLOAD FULL. PDF eBook here { https://tinyurl.com/y3nhqquc } ......................................................................................................................... 1.DOWNLOAD FULL. EPUB eBook here { https://tinyurl.com/y3nhqquc } ......................................................................................................................... 1.DOWNLOAD FULL. doc eBook here { https://tinyurl.com/y3nhqquc } ......................................................................................................................... 1.DOWNLOAD FULL. PDF eBook here { https://tinyurl.com/y3nhqquc } ......................................................................................................................... 1.DOWNLOAD FULL. EPUB eBook here { https://tinyurl.com/y3nhqquc } ......................................................................................................................... 1.DOWNLOAD FULL. doc eBook here { https://tinyurl.com/y3nhqquc } ......................................................................................................................... ......................................................................................................................... ......................................................................................................................... .............. Browse by Genre Available eBooks ......................................................................................................................... Art, Biography, Business, Chick Lit, Children's, Christian, Classics, Comics, Contemporary, CookeBOOK Crime, eeBOOK Fantasy, Fiction, Graphic Novels, Historical Fiction, History, Horror, Humor And Comedy, Manga, Memoir, Music, Mystery, Non Fiction, Paranormal, Philosophy, Poetry, Psychology, Religion, Romance, Science, Science Fiction, Self Help, Suspense, Spirituality, Sports, Thriller, Travel, Young Adult,
       Répondre 
    Voulez-vous vraiment ?  Oui  Non
    Votre message apparaîtra ici

Optimal Transport for a Computer Programmer's Point of View

  1. 1. Mathématiques - Informatique Optimal Transport From a Computer Perspective Summer School « Geometric Measure Theory and Calculus of Variations », Grenoble, June 2015 Bruno Lévy ALICE Géométrie & Lumière CENTRE INRIA Nancy Grand-Est
  2. 2. Part. 1. Goals and Motivations - Understanding Part. 2. Optimal Transport Elementary intro. Part. 3. The Semi-Discrete Case Part. 4. Understanding Going Part. 5. Concluding Words OVERVIEW
  3. 3. OVERVIEW Cédric Villani Optimal Transport Old & New Topics on Optimal Transport Yann Brenier The polar factorization theorem (Brenier Transport)
  4. 4. OVERVIEW June 2015 Institut Fourier March 2015 Bonn Febr 2015 BIRS (Canada) Febr 2015 LJLL
  5. 5. Goals and Motivations 1
  6. 6. Part. 1 Optimal Transport Measuring distances between functions
  7. 7. Part. 1 Optimal Transport Measuring distances between functions
  8. 8. Part. 1 Optimal Transport Measuring distances between functions
  9. 9. Part. 1 Optimal Transport Measuring distances between function
  10. 10. Part. 1 Optimal Transport Interpolating functions
  11. 11. Part. 1 Optimal Transport Interpolating functions
  12. 12. Part. 1 Optimal Transport Interpolating functions
  13. 13. Part. 1 Optimal Transport Interpolating functions
  14. 14. Part. 1 Optimal Transport Interpolating functions
  15. 15. Part. 1 Optimal Transport Interpolating functions
  16. 16. Part. 1 Optimal Transport Interpolating functions
  17. 17. Part. 1 Optimal Transport Interpolating functions
  18. 18. Part. 1 Optimal Transport Interpolating functions
  19. 19. Part. 1 Optimal Transport Interpolating functions
  20. 20. Part. 1 Optimal Transport Interpolating functions
  21. 21. Part. 1 Optimal Transport Interpolating functions
  22. 22. Part. 1 Optimal Transport ? T
  23. 23. Part. 1 Optimal Transport ? T distance between the shapes
  24. 24. Part. 1 Optimal Transport while preserving mass and minimizing the effort ? ? T
  25. 25. Part. 1 Optimal Transport while preserving mass and minimizing the effort ? ? T
  26. 26. Part. 1 Optimal Transport while preserving mass and minimizing the effort ? ? T
  27. 27. Part. 1 Optimal Transport OT= Yann Brenier: it can be replaced with the Monge-Ampère
  28. 28. Part. 1 Optimal Transport OT= Yann Brenier: it can be replaced with the Monge-Ampère New ways of simulating physics with a computer
  29. 29. Part. 1 Optimal Transport OT= Yann Brenier: Laplace operator is used in a PDE, it can be replaced with the Monge-Ampère New ways of simulating physics with a computer Fast Fourier Transform
  30. 30. Part. 1 Optimal Transport OT= Yann Brenier: Laplace operator is used in a PDE, it can be replaced with the Monge-Ampère New ways of simulating physics with a computer Fast Fourier Transform Fast OT algo. ???
  31. 31. Part. 1 Optimal Transport Gaspard Monge - 1784
  32. 32. Part. 1 Optimal Transport Gaspard Monge geometry and light ANR TOMMI Workshop
  33. 33. Part. 1 Optimal Transport [Castro, Merigot, Thibert 2014] Optimal transport geometry and light [Caffarelli, Kochengin, and Oliker 1999]
  34. 34. Part. 1 Optimal Transport Image Processing Barycenters / mixing textures Video-style transfer [Nicolas Bonneel, Julien Rabin, Gabriel Peyre, Hanspeter Pfister] [Nicolas Bonneel, Kalyan Sunkavalli, Sylvain Paris, Hanspeter Pfister]
  35. 35. Part. 1 Optimal Transport Optimal transport - geometry and light [Chwartzburg, Testuz, Tagliasacchi, Pauly, SIGGRAPH 2014]
  36. 36. Part. 1. Motivations Discretization of functionals involving the Monge-Ampère operator, Benamou, Carlier, Mérigot, Oudet arXiv:1408.4536 The variational formulation of the Fokker-Planck equation Jordan, Kinderlehrer and Otto SIAM J. on Mathematical Analysis Geostrophic current
  37. 37. Part. 1 Optimal Transport Optimal transport E.U.R. The data
  38. 38. Part. 1 Optimal Transport The millenium simulation project, Max Planck Institute fur Astrophysik pc/h : parsec (= 3.2 light years)
  39. 39. Part. 1 Optimal Transport Optimal transport E.U.R. The universal swimming pool
  40. 40. Part. 1 Optimal Transport Optimal transport E.U.R. let there be light ! The millenium simulation project, Max Planck Institute fur Astrophysik pc/h : parsec (= 3.2 light years)
  41. 41. Part. 1 Optimal Transport Optimal transport E.U.R. let there be light ! The millenium simulation project, Max Planck Institute fur Astrophysik pc/h : parsec (= 3.2 light years) In 2002: 5 hours of computation on a super-computer for 5000 points with a O(n3) algorithm.
  42. 42. Part. 1 Optimal Transport Optimal transport E.U.R. let there be light ! The millenium simulation project, Max Planck Institute fur Astrophysik pc/h : parsec (= 3.2 light years) In 2002: 5 hours of computation on a super-computer for 5000 points with a O(n3) algorithm. Can we do the computation with 1 000 000 points ?
  43. 43. Optimal Transport an elementary introduction 2
  44. 44. Part. 2 Optimal Transport problem (X; ) (Y; ) Two measures , such that d (x) = d (x) X Y
  45. 45. Part. 2 Optimal Transport problem A map T is a transport map between and if (T-1(B)) = (B) for any Borel subset B of Y (X; ) (Y; )
  46. 46. Part. 2 Optimal Transport problem A map T is a transport map between and if (T-1(B)) = (B) for any Borel subset B B (X; ) (Y; )
  47. 47. Part. 2 Optimal Transport problem A map T is a transport map between and if (T-1(B)) = (B) for any Borel subset B B T-1(B) (X; ) (Y; )
  48. 48. Part. 2 Optimal Transport problem A map T is a transport map between and if (T-1(B)) = (B) for any Borel subset B (or = T# the pushforward of ) (X; ) (Y; )
  49. 49. Part. 2 Optimal Transport problem problem: Find a transport map T that minimizes C(T) = X || x T(x) ||2 d (x) (X; ) (Y; )
  50. 50. Part. 2 Optimal Transport problem problem: Find a transport map T that minimizes C(T) = X || x T(x) ||2 d (x) Difficult to study If has an atom (isolated Dirac), it can only be mapped to another Dirac (T needs to be a map)
  51. 51. Part. 2 Optimal Transport problem problem: Find a transport map T that minimizes C(T) = X || x T(x) ||2 d (x) Transport from a measure concentrated on L1 onto another one concentrated on L2 and L3
  52. 52. Part. 2 Optimal Transport problem problem: Find a transport map T that minimizes C(T) = X || x T(x) ||2 d (x) Transport from a measure concentrated on L1 onto another one concentrated on L2 and L3
  53. 53. Part. 2 Optimal Transport problem problem: Find a transport map T that minimizes C(T) = X || x T(x) ||2 d (x) Transport from a measure concentrated on L1 onto another one concentrated on L2 and L3
  54. 54. Part. 2 Optimal Transport problem problem: Find a transport map T that minimizes C(T) = X || x T(x) ||2 d (x) Transport from a measure concentrated on L1 onto another one concentrated on L2 and L3 The infimum is never realized by a map, need for a relaxation
  55. 55. Part. 2 Optimal Transport Kantorovich problem: Find a transport map T that minimizes C(T) = X || x T(x) ||2 d (x) (1942): Find a measure defined on X x Y such that x in X d (x,y) = d (y) and y in Y d (x,y) = d (x) that minimizes X x Y || x y ||2 d (x,y)
  56. 56. Part. 2 Optimal Transport Kantorovich problem: Find a transport map T that minimizes C(T) = X || x T(x) ||2 d (x) Find a measure defined on X x Y such that x in X d (x,y) = d (y) and y in Y d (x,y) = d (x) that minimizes X x Y || x y ||2 d (x,y) (x,y How much sand goes from x to y
  57. 57. Part. 2 Optimal Transport Kantorovich problem: Find a transport map T that minimizes C(T) = X || x T(x) ||2 d (x) Find a measure defined on X x Y such that x in X d (x,y) = d (y) and y in Y d (x,y) = d (x) that minimizes X x Y || x y ||2 d (x,y) Everything that is transported from x
  58. 58. Part. 2 Optimal Transport Kantorovich problem: Find a transport map T that minimizes C(T) = X || x T(x) ||2 d (x) Find a measure defined on X x Y such that x in X d (x,y) = d (y) and y in Y d (x,y) = d (x) that minimizes X x Y || x y ||2 d (x,y) Everything that is transported to y
  59. 59. Part. 2 Optimal Transport Kantorovich Transport plan example 1/4 : translation of a segment
  60. 60. Part. 2 Optimal Transport Kantorovich Transport plan example 1/4 : translation of a segment
  61. 61. Part. 2 Optimal Transport Kantorovich Transport plan example 2/4 : spitting a segment
  62. 62. Part. 2 Optimal Transport Kantorovich
  63. 63. Part. 2 Optimal Transport Kantorovich
  64. 64. Part. 2 Optimal Transport Kantorovich
  65. 65. Part. 2 Optimal Transport Kantorovich Transport plan example 3/4 : splitting a Dirac into two Diracs
  66. 66. Part. 2 Optimal Transport Kantorovich Transport plan example 3/4 : splitting a Dirac into two Diracs (No transport map)
  67. 67. Part. 2 Optimal Transport Kantorovich Transport plan example 4/4 : splitting a Dirac into two segments
  68. 68. Part. 2 Optimal Transport Kantorovich Transport plan example 4/4 : splitting a Dirac into two segments (No transport map)
  69. 69. Part. 2 Optimal Transport Duality
  70. 70. Part. 2 Optimal Transport Duality Duality is easier to understand with a discrete version
  71. 71. Part. 2 Optimal Transport Duality (DMK): Min <c, > P1 = u s.t. P2 = v
  72. 72. Part. 2 Optimal Transport Duality (DMK): Min <c, > P1 = u s.t. P2 = v = 11 12 1n 22 2n mn IRmn
  73. 73. Part. 2 Optimal Transport Duality (DMK): Min <c, > P1 = u s.t. P2 = v = 11 12 1n 22 2n mn c = c11 c12 c1n c22 c2n cmn IRmnIRmn
  74. 74. Part. 2 Optimal Transport Duality (DMK): Min <c, > P1 = u s.t. P2 = v = 11 12 1n 22 2n mn c = c11 c12 c1n c22 c2n cmn cij = || xi yj ||2 IRmnIRmn
  75. 75. Part. 2 Optimal Transport Duality (DMK): Min <c, > P1 = u s.t. P2 = v 0 = 11 12 1n 22 2n mn c = c11 c12 c1n c22 c2n cmn cij = || xi yj ||2 mn X m IRmnIRmn
  76. 76. Part. 2 Optimal Transport Duality (DMK): Min <c, > P1 = u s.t. P2 = v = 11 12 1n 22 2n mn c = c11 c12 c1n c22 c2n cmn cij = || xi yj ||2 mn X n mn X m IRmnIRmn
  77. 77. Part. 2 Optimal Transport Duality (DMK): Min <c, > P1 = u s.t. P2 = v = 11 12 1n 22 2n mn c = c11 c12 c1n c22 c2n cmn IRmnIRmn mn X n mn X m
  78. 78. Part. 2 Optimal Transport Duality Consider L( , ) = <c, > - < , P1 - u> - < , P2 - v> (DMK): Min <c, > P1 = u s.t. P2 = v< u, v > denotes the dot product between u and v
  79. 79. Part. 2 Optimal Transport Duality Consider L( , ) = <c, > - < , P1 - u> - < , P2 - v> Remark: Sup[ L( , ) ] = < c, > if P1 = u and P2 = v IRm IRn (DMK): Min <c, > P1 = u s.t. P2 = v
  80. 80. Part. 2 Optimal Transport Duality Consider L( , ) = <c, > - < , P1 - u> - < , P2 - v> Remark: Sup[ L( , ) ] = < c, > if P1 = u and P2 = v IRm IRn (DMK): Min <c, > P1 = u s.t. P2 = v 0
  81. 81. Part. 2 Optimal Transport Duality Consider L( , ) = <c, > - < , P1 - u> - < , P2 - v> Remark: Sup[ L( , ) ] = < c, > if P1 = u and P2 = v IRm IRn Consider now: Inf [ Sup[ L( , ) ] ] IRm IRn (DMK): Min <c, > P1 = u s.t. P2 = v
  82. 82. Part. 2 Optimal Transport Duality Consider L( , ) = <c, > - < , P1 - u> - < , P2 - v> Remark: Sup[ L( , ) ] = < c, > if P1 = u and P2 = v IRm IRn Consider now: Inf [ Sup[ L( , ) ] ] = Inf [ < c, > ] IRm IRn P1 = u P2 = v (DMK): Min <c, > P1 = u s.t. P2 = v
  83. 83. Part. 2 Optimal Transport Duality Consider L( , ) = <c, > - < , P1 - u> - < , P2 - v> Remark: Sup[ L( , ) ] = < c, > if P1 = u and P2 = v IRm IRn Consider now: Inf [ Sup[ L( , ) ] ] = Inf [ < c, > ] (DMK) IRm IRn 0 P1 = u P2 = v (DMK): Min <c, > P1 = u s.t. P2 = v
  84. 84. Part. 2 Optimal Transport Duality IRm IRn Inf [ Sup[ <c, > - < , P1 - u> - < , P2 - v> ] ] (DMK): Min <c, > P1 = u s.t. P2 = v
  85. 85. Part. 2 Optimal Transport Duality IRm IRn 0 Inf [ Sup[ <c, > - < , P1 - u> - < , P2 - v> ] ] IRm IRn Sup[ Inf[ <c, > - < , P1 - u> - < , P2 - v> ] ] Exchange Inf Sup (DMK): Min <c, > P1 = u s.t. P2 = v
  86. 86. Part. 2 Optimal Transport Duality IRm IRn Inf [ Sup[ <c, > - < , P1 - u> - < , P2 - v> ] ] IRm IRn Sup[ Inf[ <c, > - < , P1 - u> - < , P2 - v> ] ] IRm IRn 0 Sup[ Inf[ < ,c-P1 t P2 t > + < ,u> + < , v> ] ] Exchange Inf Sup Expand/Reorder/Collect (DMK): Min <c, > P1 = u s.t. P2 = v 0
  87. 87. Part. 2 Optimal Transport Duality IRm IRn Inf [ Sup[ <c, > - < , P1 - u> - < , P2 - v> ] ] IRm IRn Sup[ Inf[ <c, > - < , P1 - u> - < , P2 - v> ] ] IRm IRn Sup[ Inf[ < ,c-P1 t P2 t > + < ,u> + < , v> ] ] Exchange Inf Sup Expand/Reorder/Collect Interpret (DMK): Min <c, > P1 = u s.t. P2 = v
  88. 88. Part. 2 Optimal Transport Duality IRm IRn Sup[ Inf[ < ,c-P1 t P2 t > + < ,u> + < , v> ] ] Interpret Sup[ < ,u> + < , v> ] IRm IRn P1 t + P2 t c (DMK): Min <c, > P1 = u s.t. P2 = v (DDMK)
  89. 89. Part. 2 Optimal Transport Duality IRm IRn Sup[ Inf[ < ,c-P1 t P2 t > + < ,u> + < , v> ] ] Interpret Sup[ < ,u> + < , v> ] IRm IRn P1 t + P2 t c i + j cij (i,j) A (DMK): Min <c, > P1 = u s.t. P2 = v (DDMK)
  90. 90. Part. 2 Optimal Transport Kantorovich dual Find a measure defined on X x Y such that x in X d (x,y) = d (x) and y in Y d (x,y) = d (x) that minimizes X x Y || x y ||2 d (x,y) Find two functions in L1( ) and in L1( ) Such that for all x,y, (x) + y ||2 that maximize X d + Y d
  91. 91. Part. 2 Optimal Transport Kantorovich dual Find a measure defined on X x Y such that x in X d (x,y) = d (x) and y in Y d (x,y) = d (x) that minimizes X x Y || x y ||2 d (x,y) Find two functions in L1( ) and in L1( ) Such that for all x,y, (x) + y ||2 that maximize X d + Y d Your point of view: Try to minimize transport cost
  92. 92. Part. 2 Optimal Transport Kantorovich dual Find a measure defined on X x Y such that x in X d (x,y) = d (x) and y in Y d (x,y) = d (x) that minimizes X x Y || x y ||2 d (x,y) Find two functions in L1( ) and in L1( ) Such that for all x,y, (x) + y ||2 that maximize X d + Y d Try to maximize transport price Your point of view: Try to minimize transport cost
  93. 93. Part. 2 Optimal Transport Kantorovich dual Find a measure defined on X x Y such that x in X d (x,y) = d (x) and y in Y d (x,y) = d (x) that minimizes X x Y || x y ||2 d (x,y) Find two functions in L1( ) and in L1( ) Such that for all x,y, (x) + y ||2 that maximize X (x)d + Y (y)d What they charge for loading at x Your point of view: Try to minimize transport cost
  94. 94. Part. 2 Optimal Transport Kantorovich dual Find a measure defined on X x Y such that x in X d (x,y) = d (x) and y in Y d (x,y) = d (x) that minimizes X x Y || x y ||2 d (x,y) Find two functions in L1( ) and in L1( ) Such that for all x,y, (x) + y ||2 that maximize X (x)d + Y (y)d What they charge for loading at x What they charge for unloading at y Your point of view: Try to minimize transport cost
  95. 95. Part. 2 Optimal Transport Kantorovich dual Find a measure defined on X x Y such that x in X d (x,y) = d (x) and y in Y d (x,y) = d (x) that minimizes X x Y || x y ||2 d (x,y) Find two functions in L1( ) and in L1( ) Such that for all x,y, (x) + y ||2 that maximize X (x)d + Y (y)d What they charge for loading at x What they charge for unloading at y Price (loading + unloading) cannot be greater than transport cost (else you do the job yourself) Your point of view: Try to minimize transport cost
  96. 96. Part. 2 Optimal Transport c-conjugate functions Find two functions in L1( ) and in L1( ) Such that for all x,y, (x) + y ||2 that maximize X (x)d + Y (y)d
  97. 97. Part. 2 Optimal Transport c-conjugate functions Find two functions in L1( ) and in L1( ) Such that for all x,y, (x) + y ||2 that maximize X (x)d + Y (y)d If we got two functions and that satisfy the constraint Then it is possible to obtain a better solution by replacing with c defined by: For all y, c(y) = inf x in X ½|| x y ||2 - (y)
  98. 98. Part. 2 Optimal Transport c-conjugate functions Find two functions in L1( ) and in L1( ) Such that for all x,y, (x) + y ||2 that maximize X (x)d + Y (y)d If we got two functions and that satisfy the constraint Then it is possible to obtain a better solution by replacing with c defined by: For all y, c(y) = inf x in X ½|| x y ||2 - (y) c is called the c-conjugate function of If there is a function such that = c then is said to be c-concave If is c-concave, then cc =
  99. 99. Part. 2 Optimal Transport c-conjugate functions Find a c-concave function that maximizes X (x)d + Y c(y)d
  100. 100. Part. 2 Optimal Transport c-conjugate functions Find a c-concave function that maximizes X (x)d + Y c(y)d is called a
  101. 101. Part. 2 Optimal Transport c-subdifferential What about our initial problem ? is called a Find a c-concave function that maximizes X (x)d + Y c(y)d
  102. 102. Part. 2 Optimal Transport c-subdifferential is called a Find a c-concave function that maximizes X (x)d + Y c(y)d
  103. 103. Part. 2 Optimal Transport c-subdifferential
  104. 104. Part. 2 Optimal Transport c-subdifferential Proof: see OTON, chap. 10.
  105. 105. Part. 2 Optimal Transport c-subdifferential Proof: see OTON, chap. 10. Heuristic argument (at the beginning of the same chapter):
  106. 106. Part. 2 Optimal Transport c-subdifferential Proof: see OTON, chap. 10. Heuristic argument (at the beginning of the same chapter):
  107. 107. Part. 2 Optimal Transport c-subdifferential Proof: see OTON, chap. 10. Heuristic argument (at the beginning of the same chapter):
  108. 108. Part. 2 Optimal Transport c-subdifferential Proof: see OTON, chap. 10. Heuristic argument (at the beginning of the same chapter):
  109. 109. Part. 2 Optimal Transport c-subdifferential Proof: see OTON, chap. 10. Heuristic argument (at the beginning of the same chapter):
  110. 110. Part. 2 Optimal Transport c-subdifferential Proof: see OTON, chap. 10. Heuristic argument (at the beginning of the same chapter):
  111. 111. Part. 2 Optimal Transport c-subdifferential Proof: see OTON, chap. 10. Heuristic argument (at the beginning of the same chapter): w
  112. 112. Part. 2 Optimal Transport c-subdifferential Proof: see OTON, chap. 10. Heuristic argument (at the beginning of the same chapter): -w
  113. 113. Part. 2 Optimal Transport c-subdifferential Find a c-concave function that maximizes X (x)d + Y c(y)d
  114. 114. Part. 2 Optimal Transport c-subdifferential Find a c-concave function that maximizes X (x)d + Y c(y)d { grad (x) with (x) := (½ x2- (x))
  115. 115. Part. 2 Optimal Transport convexity Find a c-concave function that maximizes X (x)d + Y c(y)d { grad (x) with (x) := (½ x2- (x)) is convex
  116. 116. Part. 2 Optimal Transport convexity Find a c-concave function that maximizes X (x)d + Y c(y)d { grad (x) with (x) := (½ x2- (x)) is convex
  117. 117. Part. 2 Optimal Transport convexity Find a c-concave function that maximizes X (x)d + Y c(y)d { grad (x) with (x) := (½ x2- (x)) is convex
  118. 118. Part. 2 Optimal Transport no collision If T(.) exists, then T(x) = x grad (x) = grad (½ x2- (x) ) {grad (x) Find a c-concave function that maximizes X (x)d + Y c(y)d is convex
  119. 119. Part. 2 Optimal Transport no collision If T(.) exists, then T(x) = x grad (x) = grad (½ x2- (x) ) {grad (x) Find a c-concave function that maximizes X (x)d + Y c(y)d is convex
  120. 120. Part. 2 Optimal Transport no collision If T(.) exists, then T(x) = x grad (x) = grad (½ x2- (x) ) {grad (x) Find a c-concave function that maximizes X (x)d + Y c(y)d is convex
  121. 121. Part. 2 Optimal Transport Monge-Ampere What about our initial problem ? If T(.) exists, then one can show that: T(x) = x grad (x) = grad (½ x2- (x) ) { for all borel set A, A d = T(A) (|JT|) d (change of variable) Jacobian of T (1st order derivatives) Find a c-concave function that maximizes X (x)d + Y c(y)d grad (x) with (x) := (½ x2- (x))
  122. 122. Part. 2 Optimal Transport Monge-Ampere What about our initial problem ? If T(.) exists, then one can show that: T(x) = x grad (x) = grad (½ x2- (x) ) { for all borel set A, A d = T(A) (|JT|) d = T(A) (H ) d Det. of the Hessian of (2nd order derivatives) Find a c-concave function that maximizes X (x)d + Y c(y)d grad (x) with (x) := (½ x2- (x))
  123. 123. Part. 2 Optimal Transport Monge-Ampere What about our initial problem ? T(x) = x grad (x) = grad (½ x2- (x) ) {grad (x) with (x) := (½ x2- (x)) When and have a density u and v, (H (x)). v(grad (x)) = u(x) Monge-Ampère equation for all borel set A, A d = T(A) (|JT|) d = T(A) (H ) d Find a c-concave function that maximizes X (x)d + Y c(y)d
  124. 124. Part. 2 Optimal Transport summary Find a transport map T that minimizes C(T) = X || x T(x) ||2 d (x)
  125. 125. Part. 2 Optimal Transport summary Find a transport map T that minimizes C(T) = X || x T(x) ||2 d (x) (Kantorovich formulation, dual, c-convex functions)
  126. 126. Part. 2 Optimal Transport summary Find a transport map T that minimizes C(T) = X || x T(x) ||2 d (x) (Kantorovich formulation, dual, c-convex functions) Monge-Ampère equation (When and have a density u and v resp.) Solve (H (x)). v(grad (x)) = u(x)
  127. 127. Part. 2 Optimal Transport summary Find a transport map T that minimizes C(T) = X || x T(x) ||2 d (x) (Kantorovich formulation, dual, c-convex functions) Brenier, Mc Cann, Trudinger: The optimal transport map is then given by: T(x) = grad (x) Solve (H (x)). v(grad (x)) = u(x) Monge-Ampère equation (When and have a density u and v resp.)
  128. 128. Part. 2 Optimal Transport Isoperimetric inequality For a given volume, ball is the shape that minimizes border area
  129. 129. Part. 2 Optimal Transport Isoperimetric inequality Vol(B2 n)1/n ( f n/(n-1))(n-1)/n L1 Sobolev inegality: Given f: IRn sufficiently regular Explanation in [Dario Cordero Erauquin] course notes
  130. 130. Part. 2 Optimal Transport Isoperimetric inequality n Vol(B2 n)1/n ( f n/(n-1))(n-1)/n L1 Sobolev inegality: Given f: IRn sufficiently regular Explanation in [Dario Cordero Erauquin] course notes
  131. 131. Part. 2 Optimal Transport Isoperimetric inequality Vol(B2 n)1/n ( f n/(n-1))(n-1)/n L1 Sobolev inegality: Given f: IRn sufficiently regular Explanation in [Dario Cordero Erauquin] course notes
  132. 132. Part. 2 Optimal Transport Isoperimetric inequality Vol(B2 n)1/n ( f n/(n-1))(n-1)/n L1 Sobolev inegality: Given f: IRn sufficiently regular Consider a compact set such that Vol( ) = Vol(B2 3) and f = the indicatrix function of
  133. 133. Part. 2 Optimal Transport Isoperimetric inequality Vol(B2 n)1/n ( f n/(n-1))(n-1)/n L1 Sobolev inegality: Given f: IRn sufficiently regular Consider a compact set such that Vol( ) = Vol(B2 3) and f = the indicatrix function of Vol Vol(B2 3)1/3 Vol(B2 3)2/3
  134. 134. Part. 2 Optimal Transport Isoperimetric inequality Vol(B2 n)1/n ( f n/(n-1))(n-1)/n L1 Sobolev inegality: Given f: IRn sufficiently regular Consider a compact set such that Vol( ) = Vol(B2 3) and f = the indicatrix function of Vol Vol(B2 3)1/3 Vol(B2 3)2/3 Vol = Vol 2 3)
  135. 135. Part. 2 Optimal Transport Isoperimetric inequality Vol(B2 n)1/n ( f n/(n-1))(n-1)/n L1 Sobolev inegality: a proof with OT [Gromov]
  136. 136. Part. 2 Optimal Transport Isoperimetric inequality Vol(B2 n)1/n ( f n/(n-1))(n-1)/n L1 Sobolev inegality: a proof with OT [Gromov] We suppose w.l.o.g. that f n/(n-1) = 1
  137. 137. Part. 2 Optimal Transport Isoperimetric inequality Vol(B2 n)1/n ( f n/(n-1))(n-1)/n L1 Sobolev inegality: a proof with OT [Gromov] There exists an optimal transport T = grad between f n/(n-1)(x)dx and 1B2 n/Vol(B2 n)dx We suppose w.l.o.g. that f n/(n-1) = 1
  138. 138. Part. 2 Optimal Transport Isoperimetric inequality Vol(B2 n)1/n ( f n/(n-1))(n-1)/n L1 Sobolev inegality: a proof with OT [Gromov] There exists an optimal transport T = grad between f n/(n-1)(x)dx and 1B2 n/Vol(B2 n)dx We suppose w.l.o.g. that f n/(n-1) = 1 Monge-Ampère equation: Vol(B2 n) fn/(n-1)(x) = det Hess
  139. 139. Part. 2 Optimal Transport Isoperimetric inequality Vol(B2 n)1/n ( f n/(n-1))(n-1)/n L1 Sobolev inegality: a proof with OT [Gromov] There exists an optimal transport T = grad between f n/(n-1)(x)dx and 1B2 n/Vol(B2 n)dx We suppose w.l.o.g. that f n/(n-1) = 1 Monge-Ampère equation: Vol(B2 n) fn/(n-1)(x) = det Hess Arithmetico-geometric ineq: det (H) 1/n
  140. 140. Part. 2 Optimal Transport Isoperimetric inequality Vol(B2 n)1/n ( f n/(n-1))(n-1)/n L1 Sobolev inegality: a proof with OT [Gromov] There exists an optimal transport T = grad between f n/(n-1)(x)dx and 1B2 n/Vol(B2 n)dx We suppose w.l.o.g. that f n/(n-1) = 1 Monge-Ampère equation: Vol(B2 n) fn/(n-1)(x) = det Hess Arithmetico-geometric ineq: det (H) 1/n det (Hess ) 1/n )/n
  141. 141. Part. 2 Optimal Transport Isoperimetric inequality Vol(B2 n)1/n ( f n/(n-1))(n-1)/n L1 Sobolev inegality: a proof with OT [Gromov] There exists an optimal transport T = grad between f n/(n-1)(x)dx and 1B2 n/Vol(B2 n)dx We suppose w.l.o.g. that f n/(n-1) = 1 Monge-Ampère equation: Vol(B2 n) fn/(n-1)(x) = det Hess Arithmetico-geometric ineq: det (H) 1/n det (Hess ) 1/n )/n det (Hess ) 1/n / n
  142. 142. Part. 2 Optimal Transport Isoperimetric inequality Vol(B2 n)1/n ( f n/(n-1))(n-1)/n L1 Sobolev inegality: a proof with OT [Gromov] We suppose w.l.o.g. that f n/(n-1) = 1 det (Hess ) 1/n )/n Monge-Ampère equation: Vol(B2 n) fn/(n-1)(x) = det Hess
  143. 143. Part. 2 Optimal Transport Isoperimetric inequality Vol(B2 n)1/n ( f n/(n-1))(n-1)/n L1 Sobolev inegality: a proof with OT [Gromov] We suppose w.l.o.g. that f n/(n-1) = 1 Vol(B2 n) = Vol(B2 n) f n/(n-1) = f Vol(B2 n) f 1/(n-1) det (Hess ) 1/n )/n Monge-Ampère equation: Vol(B2 n) fn/(n-1)(x) = det Hess
  144. 144. Part. 2 Optimal Transport Isoperimetric inequality Vol(B2 n)1/n ( f n/(n-1))(n-1)/n L1 Sobolev inegality: a proof with OT [Gromov] We suppose w.l.o.g. that f n/(n-1) = 1 Vol(B2 n) = Vol(B2 n) f n/(n-1) = f Vol(B2 n) f 1/(n-1) det (Hess ) 1/n )/n Monge-Ampère equation: Vol(B2 n) fn/(n-1)(x) = det Hess
  145. 145. Part. 2 Optimal Transport Isoperimetric inequality Vol(B2 n)1/n ( f n/(n-1))(n-1)/n L1 Sobolev inegality: a proof with OT [Gromov] We suppose w.l.o.g. that f n/(n-1) = 1 Vol(B2 n) = Vol(B2 n) f n/(n-1) = f Vol(B2 n) f 1/(n-1) = - grad f . grad det (Hess ) 1/n )/n Monge-Ampère equation: Vol(B2 n) fn/(n-1)(x) = det Hess
  146. 146. Part. 2 Optimal Transport Isoperimetric inequality Vol(B2 n)1/n ( f n/(n-1))(n-1)/n L1 Sobolev inegality: a proof with OT [Gromov] We suppose w.l.o.g. that f n/(n-1) = 1 Vol(B2 n) = Vol(B2 n) f n/(n-1) = f Vol(B2 n) f 1/(n-1) 1/n = - grad f . grad | grad f | (T = grad B2 n ) det (Hess ) 1/n )/n Monge-Ampère equation: Vol(B2 n) fn/(n-1)(x) = det Hess
  147. 147. Part. 2 Optimal Transport Isoperimetric inequality Vol(B2 n)1/n ( f n/(n-1))(n-1)/n L1 Sobolev inegality: a proof with OT [Gromov] We suppose w.l.o.g. that f n/(n-1) = 1 Vol(B2 n) = Vol(B2 n) f n/(n-1) = f Vol(B2 n) f 1/(n-1) = - grad f . grad | grad f | (T = grad B2 n ) Vol(B2 n)1/n det (Hess ) 1/n )/n Monge-Ampère equation: Vol(B2 n) fn/(n-1)(x) = det Hess
  148. 148. Semi-Discrete Optimal Transport 3
  149. 149. Part. 3 Optimal Transport how to program ? Continuous (X; ) (Y; )
  150. 150. Part. 3 Optimal Transport how to program ? Continuous Semi-discrete (X; ) (Y; )
  151. 151. Part. 3 Optimal Transport how to program ? Continuous Semi-discrete Discrete (X; ) (Y; )
  152. 152. Part. 3 Optimal Transport how to program ? Continuous Semi-discrete Discrete (X; ) (Y; )
  153. 153. Part. 3 Optimal Transport semi-discrete X c (x)d + Y (y)d Sup c (DMK) (X; ) (Y; )
  154. 154. Part. 3 Optimal Transport semi-discrete (X; ) (Y; ) j (yj) vj X c (x)d + Y (y)d Sup c (DMK)
  155. 155. Part. 3 Optimal Transport semi-discrete X c (x)d + Y (y)d Sup c (DMK) j (yj) vj
  156. 156. Part. 3 Optimal Transport semi-discrete X c (x)d + Y (y)d Sup c (DMK) X inf yj Y [ || x yj ||2 - (yj) ] d j (yj) vj
  157. 157. Part. 3 Optimal Transport semi-discrete X c (x)d + Y (y)d Sup c (DMK) j Lag (yj) || x yj ||2 - (yj) d X inf yj Y [ || x yj ||2 - (yj) ] d j (yj) vj
  158. 158. Part. 3 Optimal Transport semi-discrete G( ) = j Lag (yj) || x yj ||2 - (yj) d + j (yj) vj Sup c (DMK) Where: Lag (yj) = { x | || x yj ||2 (yj) < || x yj ||2 - (yj ) }
  159. 159. Part. 3 Optimal Transport semi-discrete G( ) = j Lag (yj) || x yj ||2 - (yj) d + j (yj) vj Sup c (DMK) Where: Lag (yj) = { x | || x yj ||2 (yj) < || x yj ||2 - (yj ) } Laguerre diagram of the yj (with the L2 cost || x y ||2 used here, Power diagram)
  160. 160. Part. 3 Optimal Transport semi-discrete G( ) = j Lag (yj) || x yj ||2 - (yj) d + j (yj) vj Sup c (DMK) Where: Lag (yj) = { x | || x yj ||2 (yj) < || x yj ||2 - (yj ) } Laguerre diagram of the yj (with the L2 cost || x y ||2 used here, Power diagram) Weight of yj in the power diagram
  161. 161. Part. 3 Optimal Transport semi-discrete G( ) = j Lag (yj) || x yj ||2 - (yj) d + j (yj) vj Sup c (DMK) Where: Lag (yj) = { x | || x yj ||2 (yj) < || x yj ||2 - (yj ) } Laguerre diagram of the yj (with the L2 cost || x y ||2 used here, Power diagram) Weight of yj in the power diagram is determined by the weight vector [ (y1) (y2 (ym)]
  162. 162. Part. 3 Optimal Transport semi-discrete G( ) = j Lag (yj) || x yj ||2 - (yj) d + j (yj) vj Sup c (DMK) Where: Lag (yj) = { x | || x yj ||2 (yj) < || x yj ||2 - (yj ) } Laguerre diagram of the yj (with the L2 cost || x y ||2 used here, Power diagram) Weight of yj in the power diagram is determined by the weight vector [ (y1) (y2 (ym)] For all weight vector, is c-concave
  163. 163. Voronoi diagram: Vor(xi) = { x | d2(x,xi) < d2(x,xj) } Part. 3 Power Diagrams
  164. 164. Power diagram: Pow(xi) = { x | d2(x,xi) i < d2(x,xj) j } Voronoi diagram: Vor(xi) = { x | d2(x,xi) < d2(x,xj) } Part. 3 Power Diagrams
  165. 165. Part. 3 Power Diagrams
  166. 166. Part. 3 Optimal Transport Theorem: (direct consequence of MK duality alternative proof in [Aurenhammer, Hoffmann, Aronov 98] ): Given a measure with density, a set of points (yj), a set of positive coefficients vj such that vj d (x), it is possible to find the weights W = [ (y1) (y2 (ym)] such that the map TS W is the unique optimal transport map between and = vj (yj)
  167. 167. Part. 3 Optimal Transport Theorem: (direct consequence of MK duality alternative proof in [Aurenhammer, Hoffmann, Aronov 98] ): Given a measure with density, a set of points (yj), a set of positive coefficients vj such that vj d (x), it is possible to find the weights W = [ (y1) (y2 (ym)] such that the map TS W is the unique optimal transport map between and = vj (yj) Proof: G( ) = j Lag (yj) || x yj ||2 - (yj) d + j (yj) vj Is a concave function of the weight vector [ (y1) (y2 (ym)]
  168. 168. Part. 3 Optimal Transport Theorem: (direct consequence of MK duality alternative proof in [Aurenhammer, Hoffmann, Aronov 98] ): Given a measure with density, a set of points (yj), a set of positive coefficients vj such that vj d (x), it is possible to find the weights W = [ (y1) (y2 (ym)] such that the map TS W is the unique optimal transport map between and = vj (yj) Proof: G( ) = j Lag (yj) || x yj ||2 - (yj) d + j (yj) vj Is a concave function of the weight vector [ (y1) (y2 (ym)]
  169. 169. Part. 3 Optimal Transport the AHA paper Idea of the proof Consider the function fT(W) = (|| x T(x) ||2 (T(X)))d (x) The (unknown) weights W = [ (y1) (y2 (ym)]
  170. 170. Part. 3 Optimal Transport the AHA paper Idea of the proof Consider the function fT(W) = (|| x T(x) ||2 (T(X)))d (x) T : an arbitrary but fixed assignment.
  171. 171. Part. 3 Optimal Transport the AHA paper Idea of the proof Consider the function fT(W) = (|| x T(x) ||2 (T(X)))d (x) T : an arbitrary but fixed assignment.
  172. 172. Part. 3 Optimal Transport the AHA paper Idea of the proof Consider the function fT(W) = (|| x T(x) ||2 (T(X)))d (x) T : an arbitrary but fixed assignment.
  173. 173. Part. 3 Optimal Transport the AHA paper Idea of the proof Consider the function fT(W) = (|| x T(x) ||2 (T(X)))d (x) T : an arbitrary but fixed assignment.
  174. 174. Part. 3 Optimal Transport the AHA paper Idea of the proof Consider the function fT(W) = (|| x T(x) ||2 (T(X)))d (x) W fT(W) fT(W)
  175. 175. Part. 3 Optimal Transport the AHA paper Idea of the proof Consider the function fT(W) = (|| x T(x) ||2 (T(X)))d (x) W fT(W) fT(W)
  176. 176. Part. 3 Optimal Transport the AHA paper Idea of the proof Consider the function fT(W) = (|| x T(x) ||2 (T(X)))d (x) fT(W) is linear in W fT(W) W fT(W) Fixed T
  177. 177. Part. 3 Optimal Transport the AHA paper Idea of the proof Consider the function fT(W) = (|| x T(x) ||2 (T(X)))d (x) fT(W) is linear in W f TW (W) : defined by power diagram fT(W) fixed W W
  178. 178. Part. 3 Optimal Transport the AHA paper Idea of the proof Consider the function fT(W) = (|| x T(x) ||2 (T(X)))d (x) fT(W) is linear in W f TW (W) = minT fT(W) fT(W) fixed W W
  179. 179. Part. 3 Optimal Transport the AHA paper Idea of the proof Consider the function fT(W) = (|| x T(x) ||2 (T(X)))d (x) fT(W) is linear in W TW (W) is concave !! (because its graph is the lower enveloppe of linear functions) fT(W) W fT(W) f TW (W) = minT fT(W)
  180. 180. Part. 3 Optimal Transport the AHA paper Idea of the proof Consider the function fT(W) = (|| x T(x) ||2 (T(X)))d (x) fT(W) is linear in W TW (W) is concave (because its graph is the lower enveloppe of linear functions) fT(W) W fT(W) f TW (W) = minT fT(W) Consider g(W) = f TW vj j
  181. 181. Part. 3 Optimal Transport the AHA paper fT(W) W fT(W) f TW (W) = minT fT(W) Consider g(W) = f TW vj j vj - pow(yj)|| x yj ||2d (x) and g is concave.j = Idea of the proof Consider the function fT(W) = (|| x T(x) ||2 (T(X)))d (x) fT(W) is linear in W TW (W) is concave (because its graph is the lower enveloppe of linear functions)
  182. 182. Part. 3 Optimal Transport the algorithm Semi-discrete OT Summary: X c (x)d + Y (y)d Sup c (DMK) G( ) =
  183. 183. Part. 3 Optimal Transport the algorithm Semi-discrete OT Summary: G( ) = g(W) = j Lag (yj) || x yj ||2 - (yj) d + j (yj) vj is concave X c (x)d + Y (y)d Sup c (DMK) G( ) =
  184. 184. Part. 3 Optimal Transport the algorithm Semi-discrete OT Summary: G( ) = g(W) = j Lag (yj) || x yj ||2 - (yj) d + j (yj) vj is concave vj - pow(yj)|| x yj ||2d (x) (= 0 at the maximum)G / j = X c (x)d + Y (y)d Sup c (DMK) G( ) =
  185. 185. Part. 3 Optimal Transport the algorithm Semi-discrete OT Summary: G( ) = g(W) = j Lag (yj) || x yj ||2 - (yj) d + j (yj) vj is concave vj - pow(yj)|| x yj ||2d (x) (= 0 at the maximum)G / j = X c (x)d + Y (y)d Sup c (DMK) G( ) = Desired mass at yj Mass transported to yj
  186. 186. Part. 3 Optimal Transport the algorithm The [AHA] paper summary: The optimal weights minimize a convex function The gradient of this convex function is easy to compute Note: the weight w(s) correspond to the Kantorovich potential (x) Monge- The algorithm: Input: two tetrahedral meshes M1 and M2 Output: a morphing between M1 and M2
  187. 187. Part. 3 Optimal Transport the algorithm The [AHA] paper summary: The optimal weights minimize a convex function The gradient of this convex function is easy to compute Note: the weight w(s) correspond to the Kantorovich potential (x) Monge- The algorithm: Input: two tetrahedral meshes M1 and M2 Output: a morphing between M1 and M2 Step 1: sample M2 with N points (s1 sN)
  188. 188. Part. 3 Optimal Transport the algorithm The [AHA] paper summary: The optimal weights minimize a convex function The gradient of this convex function is easy to compute Note: the weight w(s) correspond to the Kantorovich potential (x) Monge- The algorithm: Input: two tetrahedral meshes M1 and M2 Output: a morphing between M1 and M2 Step 1: sample M2 with N points (s1 sN) Step 2: initialize the weights (w1 wN) = (0 0)
  189. 189. Part. 3 Optimal Transport the algorithm The [AHA] paper summary: The optimal weights minimize a convex function The gradient of this convex function is easy to compute Note: the weight w(s) correspond to the Kantorovich potential (x) Monge- The algorithm: Input: two tetrahedral meshes M1 and M2 Output: a morphing between M1 and M2 Step 1: sample M2 with N points (s1 sN) Step 2: initialize the weights (w1 wN Step 3: minimize g(w1 wN) with a quasi-Newton algorithm:
  190. 190. Part. 3 Optimal Transport the algorithm The [AHA] paper summary: The optimal weights minimize a convex function The gradient of this convex function is easy to compute Note: the weight w(s) correspond to the Kantorovich potential (x) Monge- The algorithm: Input: two tetrahedral meshes M1 and M2 Output: a morphing between M1 and M2 Step 1: sample M2 with N points (s1 sN) Step 2: initialize the weights (w1 wN Step 3: minimize g(w1 wN) with a quasi-Newton algorithm: For each iterate (s1 sN)(k):
  191. 191. Part. 3 Optimal Transport the algorithm The [AHA] paper summary: The optimal weights minimize a convex function The gradient of this convex function is easy to compute Note: the weight w(s) correspond to the Kantorovich potential (x) Monge- The algorithm: Input: two tetrahedral meshes M1 and M2 Output: a morphing between M1 and M2 Step 1: sample M2 with N points (s1 sN) Step 2: initialize the weights (w1 wN) = (0 0) Step 3: minimize g(w1 wN) with a quasi-Newton algorithm: For each iterate (s1 sN)(k): Compute Pow( (wi, si) M1 [Nivoliers, L 2014, Curves and Surfaces]
  192. 192. Part. 3 Optimal Transport the algorithm Compute Pow( (wi, si) M1 [Nivoliers, L 2014, Curves and Surfaces] Implementation in GEOGRAM (http://alice.loria.fr/software/geogram
  193. 193. Part. 3 Optimal Transport the algorithm The [AHA] paper summary: The optimal weights minimize a convex function The gradient of this convex function is easy to compute Note: the weight w(s) correspond to the Kantorovich potential (x) Monge- The algorithm: Input: two tetrahedral meshes M1 and M2 Output: a morphing between M1 and M2 Step 1: sample M2 with N points (s1 sN) Step 2: initialize the weights (w1 wN Step 3: minimize g(w1 wN) with a quasi-Newton algorithm: For each iterate (s1 sN)(k): Compute Pow( (wi, si) M1 [L 2014, Curves and Surfaces] Compute g and grad g
  194. 194. Part. 3 Optimal Transport the algorithm The [AHA] paper summary: The optimal weights minimize a convex function The gradient of this convex function is easy to compute Note: the weight w(s) correspond to the Kantorovich potential (x) Monge- The algorithm: Input: two tetrahedral meshes M1 and M2 Output: a morphing between M1 and M2 Step 1: sample M2 with N points (s1 sN) Step 2: initialize the weights (w1 wN Step 3: minimize g(w1 wN) with a quasi-Newton algorithm: For each iterate (s1 sN)(k): Compute Pow( (wi, si) M1 [L 2014, Curves and Surfaces] Compute g and grad g + Multilevel version [Merigot 2011] (2D), [L 2014 arXiv, M2AN 2015] (3D)
  195. 195. Part. 3 Optimal Transport the algorithm The [AHA] paper summary: The optimal weights minimize a convex function The gradient of this convex function is easy to compute Note: the weight w(s) correspond to the Kantorovich potential (x) Monge- The algorithm: Summary: The algorithm computes the weights wi such that the power cells associated with the Diracs correspond to the preimages of the Diracs.
  196. 196. Part. 3 Optimal Transport the algorithm The [AHA] paper summary: The optimal weights minimize a convex function The gradient of this convex function is easy to compute Note: the weight w(s) correspond to the Kantorovich potential (x) Monge- The algorithm: Summary: The algorithm computes the weights wi such that the power cells associated with the Diracs correspond to the preimages of the Diracs.
  197. 197. Part. 3 Optimal Transport the algorithm The [AHA] paper summary: The optimal weights minimize a convex function The gradient of this convex function is easy to compute Note: the weight w(s) correspond to the Kantorovich potential (x) Monge- The algorithm: Summary: The algorithm computes the weights wi such that the power cells associated with the Diracs correspond to the preimages of the Diracs.
  198. 198. Part. 3 Optimal Transport the algorithm The [AHA] paper summary: The optimal weights minimize a convex function The gradient of this convex function is easy to compute Note: the weight w(s) correspond to the Kantorovich potential (x) Monge- The algorithm: Summary: The algorithm computes the weights wi such that the power cells associated with the Diracs correspond to the preimages of the Diracs.
  199. 199. Part. 3 Optimal Transport the algorithm The [AHA] paper summary: The optimal weights minimize a convex function The gradient of this convex function is easy to compute Note: the weight w(s) correspond to the Kantorovich potential (x) Monge- The algorithm: Summary: The algorithm computes the weights wi such that the power cells associated with the Diracs correspond to the preimages of the Diracs.
  200. 200. Understanding Going 4
  201. 201. Part. 4 Optimal Transport ??? Wait a minute: This means that one can move (and possibly deform) a power diagram simply by changing the weights ?
  202. 202. Part. 4 Optimal Transport ??? Wait a minute: This means that one can move (and possibly deform) a power diagram simply by changing the weights ? Reminder: Power diagram in 2d = intersection between Voronoi diagram in 3d and IR2 hi = sqrt(wmax wi) Height of point i Weight of point i
  203. 203. Part. 4 Power Diagrams & Transport
  204. 204. Part. 4 Power Diagrams & Transport
  205. 205. Part. 4 Power Diagrams & Transport
  206. 206. Part. 4 Power Diagrams & Transport
  207. 207. Part. 4 Power Diagrams & Transport
  208. 208. Part. 4 Power Diagrams & Transport
  209. 209. Part. 4 Power Diagrams & Transport hi
  210. 210. Part. 4 Power Diagrams & Transport
  211. 211. Part. 4 Power Diagrams & Transport
  212. 212. Part. 4 Power Diagrams & Transport
  213. 213. Part. 4 Power Diagrams & Transport
  214. 214. Part. 4 Optimal Transport still lost ? 2D examples Numerical Experiment: Simple translation
  215. 215. Part. 4 Power Diagrams & Transport
  216. 216. Part. 4 Power Diagrams and Transport
  217. 217. Part. 4 Optimal Transport 2D examples Numerical Experiment: Splitting a disk
  218. 218. Part. 4 Optimal Transport 2D examples Numerical Experiment: A disk becomes two disks
  219. 219. Part. 4 Optimal Transport 3D examples Numerical Experiment: A sphere becomes a cube
  220. 220. Part. 4 Optimal Transport 3D examples Numerical Experiment: A sphere becomes two spheres
  221. 221. Part. 4 Optimal Transport 3D examples Numerical Experiment: Armadillo to sphere
  222. 222. Part. 4 Optimal Transport 3D examples Numerical Experiment: Other examples
  223. 223. Part. 4 Optimal Transport 3D examples Numerical Experiment: Varying density
  224. 224. Part. 4 Optimal Transport 3D examples Numerical Experiment: Performances
  225. 225. Part. 4 Optimal Transport 3D examples Numerical Experiment: Performances Note that a few years ago, several hours of supercomputer time were needed for computing OT with a few thousand Dirac masses, with a combinatorial algorithm in O(n3)
  226. 226. Part. 4 Optimal Transport 3D examples Numerical Experiment: Performances Note that a few years ago, several hours of supercomputer time were needed for computing OT with a few thousand Dirac masses, with a combinatorial algorithm in O(n3) With the semi-discrete algorithm, it takes less than 10 seconds on my laptop
  227. 227. Part. 4 Optimal Transport 3D examples Numerical Experiment: Performances Note that a few years ago, several hours of supercomputer time were needed for computing OT with a few thousand Dirac masses, with a combinatorial algorithm in O(n3) In the semi-discrete setting, my 3D version of multigrid algorithm computes OT for 1 million Dirac masses in less than 1 hour on a laptop PC
  228. 228. Part. 4 Optimal Transport 3D examples Numerical Experiment: Performances Note that a few years ago, several hours of supercomputer time were needed for computing OT with a few thousand Dirac masses, with a combinatorial algorithm in O(n3) In the semi-discrete setting, my 3D version of multigrid algorithm computes OT for 1 million Dirac masses in less than 1 hour on a laptop PC Even much faster convergence can probably be reached with a true Newton
  229. 229. Conclusions Open questions * Connections with physics, Legendre transform and entropy ? [Cuturi & Peyré] regularized discrete optimal transport why does it work ? Hint 1: Minimum action principle subject to conservation laws * More continuous numerical algorithms ? [Benamou & Brenier] fluid dynamics point of view very elegant, but 4D problem !! FEM-type adaptive discretization of the subdifferential (graph of T) ? * Can we characterize OT in other semi-discrete settings ? measures supported on unions of spheres piecewise linear densities * Connections with computational geometry ? Singularity set [Figalli] = set of points where T is discontinuous Voronoi diagrams
  230. 230. Conclusions - References Some references (that this presentation is based on) A Multiscale Approach to Optimal Transport, Quentin Mérigot, Computer Graphics Forum, 2011 Variational Principles for Minkowski Type Problems, Discrete Optimal Transport, and Discrete Monge-Ampere Equations Xianfeng Gu, Feng Luo, Jian Sun, S.-T. Yau, ArXiv 2013 Minkowski-type theorems and least-squares clustering AHA! (Aurenhammer, Hoffmann, and Aronov), SIAM J. on math. ana. 1998 Topics on Optimal Transportation, 2003 Optimal Transport Old and New, 2008 Cédric Villani
  231. 231. Conclusions - References Other references (that I need to read and understand) Polar factorization and monotone rearrangement of vector-valued functions Yann Brenier, Comm. On Pure and Applied Mathematics, June 1991 A computational fluid mechanics solution of the Monge-Kantorovich mass transfer problem, J.-D. Benamou, Y. Brenier, Numer. Math. 84 (2000), pp. 375-393 Pogorelov, Alexandrov Gradient maps, Minkovsky problem (older than AHA paper, some overlap, in slightly different context, formalism used by Gu & Yau) Rockafeller Convex optimization Theorem to switch inf(sup()) sup(inf()) with convex functions (used to justify Kantorovich duality) New textbook: Filippo Santambrogio Optimal Transport for Applied Mathematician, Calculus of Variations, PDEs and Modeling Jan 15, 2015
  232. 232. Online resources All the sourcecode/documentation available from: alice.loria.fr/software/geogram Computes semi-discrete OT in 3D Scales up to millions Dirac masses on a laptop L., A numerical algorithm for semi-discrete L2 OT in 3D, ESAIM Math. Modeling and Analysis, accepted (draft: http://arxiv.org/abs/1409.1279 <= to be fixed: bug in MA equation in this version, fixed in M2AN journal version)
  233. 233. Acknowledgements Funding: European Research Council french GOODSHAPE ERC-StG-205693 VORPALINE ERC-PoC-334829 ANR MORPHO, ANR BECASIM Quentin Merigot, Yann Brenier, Boris Thibert, Emmanuel Maitre, Jean-David Benamou, Filippo Santambrogio, Edouard Oudet, Hervé Pajot. ANR TOMMI, ANR GEOMETRYA

×