SlideShare a Scribd company logo
Voronoi diagrams in information geometry 
 Statisti 
al Voronoi diagrams and their appli 
ations  
Frank NIELSEN 
É 
ole Polyte 
hnique 
Sony CSL 
e-mail: Frank.Nielsen a 
m.org 
MaxEnt 2014 

 
2014 Frank NIELSEN 5793b870 1/57
Outline of the tutorial 
1. Eu 
lidean Voronoi diagrams 
2. Dis 
riminant analysis, Mahalanobis Voronoi diagrams, and 
anisotropi 
Voronoi diagrams 
3. Fisher-Hotelling-Rao Voronoi diagrams (Riemannian 
urved geometries) 
4. Kullba 
k-Leibler Voronoi diagrams (Bregman Voronoi 
diagrams, dually at geometries) 
5. Bayes' error, Cherno information and statisti 
al Voronoi 
diagrams 

 
2014 Frank NIELSEN 5793b870 2/57
Eu 
lidean (ordinary) Voronoi diagrams 
P = {P1, ..., Pn} : n distin 
t point generators in Eu 
lidean spa 
e Ed 
=1Bi+(Pi , Pj ) 
V (Pi ) = {X : DE (Pi ,X) ≤ DE (Pj ,X), ∀j6= i} = ∩ni 
DE (P,Q) = k(P) − (Q)k2 = 
qPd 
i=1(i (P) − i (Q))2 
(P) = p : Cartesian 
oordinate system with j (Pi ) = p(j) 
i . 
Bise 
tors Bi(P,Q) = {X : DE (P,X) ≤ DE (Q,X)} : hyperplanes 
Voronoi diagram = 
ell 
omplex V (Pi )'s with their fa 
es 
⇒ Many appli 
ations : 
rystal growth, 
odebook/quantization, 
mole 
ule interfa 
es/do 
king, motion planning, et 
. 

 
2014 Frank NIELSEN 5793b870 3/57
Voronoi diagrams and dual Delaunay simpli 
ial 
omplex 
⇒ Empty sphere property, max min angle triangulation, et 
⇒ General position : no (d + 2) points 
ospheri 
al 
⇒ Voronoi  dual Delaunay triangulation 
⇒ Bise 
tor Bi(P,Q) perpendi 
ular ⊥ to segment [PQ] 

 
2014 Frank NIELSEN 5793b870 4/57
Minimum spanning tree ⊂ Delaunay triangulation 
All edges of the Eu 
lidean MST are Delaunay edges 
→ Prim's greedy algorithm in Delaunay graph in O(n log n) 
Convex hull : boundary @ of the Delaunay triangulation 

 
2014 Frank NIELSEN 5793b870 5/57
Order k Voronoi diagrams 
All subsets of size k, Pk = 
P 
k 
 
= {K1, ...,KN} with N = 
n 
k 
 
. 
partition the spa 
e into non-empty k-order Voronoi 
ells : 
Vork (Ki ) = {x : ∀q ∈ Ki , ∀r ∈ PKi , D(x, q) ≤ D(x, r )} 
Combinatorial 
omplexity not yet settled ! ! ! (k-sets/levels) 
≡ proje 
tion of k-levels of arrangements of hyperplanes in Rd+1. 

 
2014 Frank NIELSEN 5793b870 6/57
Farthest order Voronoi and Minimum En 
losing Ball 
Cir 
um 
enter/minmax 
enter on the farthest Voronoi diagram. 
Non-dierentiable optim. at the farthest Voronoi [16℄ 

 
2014 Frank NIELSEN 5793b870 7/57
Voronoi  Delaunay : Complexity and algorithms 
◮ Complexity : (n⌈ d 
2 ⌉) (quadrati 
in 3D) 
Moment 
urve : t7→ (t, t2, .., td ), et 
. 
◮ Constru 
tion : (n log n + n⌈ d 
2 ⌉), output-sensitive algorithms 

(n log n + f ), not yet optimal output-sensitive 
algorithms. 
◮ Voronoi diagram : VorD(P) = Vorf (D(·,·))(P) for a stri 
tly 
monotoni 
ally in 
reasing fun 
tion f (·) 
Vor. Eu 
lidean ≡ Vor. squared Eu 
lidean ⊂ Vor. Bregman 
◮ Geometri 
toolbox : spa 
e of spheres (polarity, orthogonality 
between spheres, et 
.) 

 
2014 Frank NIELSEN 5793b870 8/57
Multiple Hypothesis Testing [9, 10℄ 
Given a random variable X with n hypothesis 
H1 : X ∼ P1 
: ... 
Hn : X ∼ Pn 
de 
ide for a IID sample x1, ..., xm ∼ X whi 
h hypothesis holds true ? 
Pm 
orre 
t = 1 − Pm 
error 
Asymptoti 
regime : Error exponent  
lim 
m→∞ 
− 
1 
m 
log Pm 
e =  

 
2014 Frank NIELSEN 5793b870 9/57
Bayesian hypothesis testing 
◮ Prior probabilities : wi = P(X ∼ Pi )  0 (with 
Pn 
i=1 wi = 1) 
◮ Conditional probabilities : P(X = x|X ∼ Pi ). 
P(X = x) = 
Xn 
i=1 
P(X ∼ Pi )P(X = x|X ∼ Pi ) = 
Xn 
i=1 
wiP(X|Pi ) 
◮ Cost design matrix [cij ]= with ci ,j = 
ost of de 
iding Hi when 
in fa 
t Hj is true 
◮ pi ,j (u(x)) = probability of making this de 
ision using rule u. 

 
2014 Frank NIELSEN 5793b870 10/57
Bayesian dete 
tor : MAP rule 
◮ Minimize the expe 
ted 
ost : 
EX [c(r (x))], c(r (x)) = 
X 
i 
 
wi 
X 
j6=i 
 
 
ci ,jpi ,j (r (x)) 
◮ Spe 
ial 
ase : Probability of error Pe : ci ,i = 0 and ci ,j = 1 for 
i6= j : 
Pe = EX 
 
 
X 
i 
 
wi 
X 
j6=i 
 
 
pi ,j (r (x)) 
 
 
◮ Maximum a posteriori probability (MAP) rule : 
map(x) = argmaxi∈{1,...,n} wipi (x) 
where pi (x) = P(X = x|X ∼ Pi ) are the 
onditional 
probabilities. 
→ MAP Bayesian dete 
tor minimizes Pe over all rules [7℄ 

 
2014 Frank NIELSEN 5793b870 11/57
MVNs, MAP rule, and Mahalanobis distan 
e (1930) 
MultiVariate Normals 
(MVNs) 
◮ Say, X1 ∼ N(μ1,) and X2 ∼ N(μ2,) 
◮ Probability of error (mis 
lassi 
ation) for w1 = w2 = 12 
: 
Pe = 
1 
2 
(1 − TV(P1, P2)) =  
 
− 
1 
2 
 
(P1, P2) 
(·) : standard normal distribution fun 
tion 
◮ Mahalanobis metri 
distan 
e (·, ·) : 
2(X1,X2) = (μ1 − μ2)⊤−1(μ1 − μ2) = μ⊤−1μ 
◮ Measure the density overlapping (the greater, the lesser) 
◮ generalize Eu 
lidean distan 
e ( = I ) 
◮ 2 : only symmetri 
Bregman divergen 
e. 
⇒ easy to approximate P˜e experimentally 

 
2014 Frank NIELSEN 5793b870 12/57
Mahalanobis Voronoi diagrams 
MHT for isotropi 
MVNs with uniform weight wi = 
1 
n 
Cholesky de 
omposition  = LL⊤ 
2(X1,X2) = (μ1 − μ2)⊤−1(μ1 − μ2), 
= (μ1 − μ2)⊤(L−1)⊤L−1(μ1 − μ1), 
= (L−1μ1 − L−1μ2)⊤(L−1μ1 − L−1μ2) 
2(X1,X2) = DE (L−1μ1, L−1μ2) 
Mahalanobis Voronoi diagram : 
◮ Cholesky de 
omposition :  = LL⊤ 
◮ Map P = {p1, ..., pn} to P′ = {p′ 1 
, ..., p′n 
i = L−1x. 
} with p′ 
◮ Ordinary Voronoi diagram VorE (P′) 
◮ Map VorE (P′) to Vor(P) = LVorE (P′). 

 
2014 Frank NIELSEN 5793b870 13/57
Mahalanobis Voronoi diagrams 
 a 
ount for both 
orrelation and dimension (feature) s 
aling 
Dual stru 
ture ≡ anisotropi 
Delaunay triangulation 
⇒ empty 
ir 
umellipse property 
Re 
all : MAP rule from Voronoi diagram 

 
2014 Frank NIELSEN 5793b870 14/57
Anisotropi 
Voronoi diagrams [6℄ (MHT for MVNs) 
Classi 
ation : Generator Xi ∼ N(μi ,i ), wi = 1 
n . 
Dis 
riminant fun 
tions are quadrati 
bise 
tors 
Orphans, islands, dual not a triangulation 
(need wedged/visibility 
onditions) 
Appli 
ations : Crystal growth in elds (not Riemannian smooth 
metri 
) 

 
2014 Frank NIELSEN 5793b870 15/57
Statisti 
s : Estimators 
Given x1, ..., xn IID observations (from a population), estimate the 
underlying distribution X ∼ p. 
◮ empiri 
al CDF : ˆ Fn(x) = 1 
n 
P 
i I(−∞,x)(xi ). 
Glivenko-Cantelli theorem : supx | ˆ Fn(x) − F(x)| → 0 a.s. 
◮ Fisher approa 
h : Density p belongs to a parametri 
family 
p(x|), D = #parameters (order) 
◮ Method of moments : Mat 
h distribution moments with 
sample moments : 
EX [Xl ] = 
1 
n 
Xn 
i=1 
xl 
i , 
⇒ any D independent equations yields an estimator 
◮ Fisher Maximum (log-)Likelihood Estimator MLE : 
max 
 
l (; x1, ..., xn) = max 
 
Yn 
i=1 
p(xi ; ) 

 
2014 Frank NIELSEN 5793b870 16/57
Estimation : Varian 
e Lower Bound 
Whi 
h estimator ˆ n shall we 
hoose ? 
Estimator is also a random variable on a random ve 
tor 
◮ Consistent : limn→∞ ˆn →  
◮ Unbiased : E[ˆn] −  = 0 
◮ Minimum square error (MSE) : E[(ˆ − )2] = B[ˆ]2 + V[ˆ] 
◮ Fisher information matrix : 
I () = 
 
Ii ,j () = E 
 
@ 
@i 
log p(x|) 
@ 
@j 
 
log p(x|) 
◮ Fré 
het Darmois Cramér-Rao lower bound for unbiased 
estimators : 
V[ ˆ n]  
1 
n 
I−1() 
◮ E 
ient : estimator rea 
hes the Cramèr-Rao lower bound 

 
2014 Frank NIELSEN 5793b870 17/57
Su 
ient statisti 
s and exponential families 
◮ Su 
ient statisti 
t(x) : 
P(x|t(x) = t, ) = P(x|t(x) = t) 
All information for inferen 
e 
ontained in t(x). (6= an 
illary) 
◮ For univariate Gaussians, D = 2 : t1(x) = x and t2(x) = x2. 
ˆμ = ¯x = 
1 
n 
X 
i 
xi , ˆ2 = 
1 
n 
X 
i 
(xi −¯x)2 = 
  
1 
n 
X 
i 
x2 
i 
! 
−(¯x)2 
◮ Exponential families have nite dimensional su 
ient 
statisti 
s (Koopman) : Redu 
e n data to D statisti 
s. 
∀x ∈ X, P(x|) = exp(⊤t(x) − F() + k(x)) 
F(·) : log-normalizer/ 
umulant/partition fun 
tion 
k(x) : auxiliary term for 
arrier measure 

 
2014 Frank NIELSEN 5793b870 18/57
Exponential families : Fisher information and MLE 
Common distributions are exponential families : 
Poisson, Gaussians, Gamma, Beta, Diri 
hlet, et 
. 
◮ Fisher information : 
I () = ∇2F() 
◮ MLE for an exponential family : 
∇F() = 
1 
n 
X 
i 
t(Xi ) =  
MLE exists i. 
 = ∇F() ∈ int(CH(X)) 

 
2014 Frank NIELSEN 5793b870 19/57
Population spa 
e : Hotelling (1930) [5℄  Rao (1945) [25℄ 
Birth of dierential-geometri 
methods in statisti 
s. 
◮ Fisher information matrix (positive denite) 
an be used as a 
(smooth) Riemannian metri 
tensor. 
◮ Distan 
e between two populations indexed by 1 and 2 : 
Riemannian distan 
e (metri 
length) 
◮ Fisher-Hotelling-Rao (FHR) geodesi 
distan 
e used in 
lassi 
ation : Find the 
losest population to a given 
population 
◮ Used in tests of signi 
an 
e (null versus alternative 
hypothesis), power of a test : P(reje 
t H0|H0 is false ) 
(surfa 
es in population spa 
es) 

 
2014 Frank NIELSEN 5793b870 20/57
Rao's distan 
e (1945, introdu 
ed by Hotelling 1930 [5℄) 
◮ Innitesimal length element : 
ds2 = 
X 
ij 
gij ()didj = dT I ()d 
◮ Geodesi 
and distan 
e are hard to expli 
itly 
al 
ulate : 
(p(x; 1), p(x; 2)) = min 
(s) 
(0)=1 
(1)=2 
Z 1 
0 
s 
d 
ds 
T 
I () 
d 
ds 
ds 
◮ Metri 
property of , many tools [1℄ : Riemannian Log/Exp 
tangent/manifold mapping 

 
2014 Frank NIELSEN 5793b870 21/57
Fisher Rao Hotelling Voronoi : Riemannian Voronoi diagrams 
◮ Lo 
ation-s 
ale 2D families have non-positive 
urvature 
(Hotelling, 1930) : FHR Voronoi diagrams amount to 
hyperboli 
Voronoi diagrams or Eu 
lidean diagrams 
(lo 
ation families only like isotropi 
Gaussians) 
◮ Multinomial family has spheri 
al geometry on the positive 
orthant : Spheri 
al Voronoi diagram (stereographi 
proje 
tion ∝ Eu 
lidean Voronoi diagrams) 
◮ Arbitrary families p(x|) : Geodesi 
s not in 
losed forms → 
limited 
omputational framework in pra 
ti 
e... 

 
2014 Frank NIELSEN 5793b870 22/57
Hyperboli 
Voronoi diagrams [19, 22℄ 
◮ In Klein disk, the hyperboli 
Voronoi diagram amounts to a 
lipped ane Voronoi diagram, or a 
lipped power 
diagram. E 
ient 
lipping algorithm [2℄. 
◮ Convert to other models of hyperboli 
geometry : Poin 
aré 
disk, upper half spa 
e, hyperboloid, Beltrami hemisphere, et 
. 
◮ Conformal (good for vizualizing) versus non- 
onformal 
(good for 
omputing) models. ( 
onformal metri 
G(x) = (x)I is s 
aled identity metri 
) 

 
2014 Frank NIELSEN 5793b870 23/57
Hyperboli 
Voronoi diagrams [19, 22℄ 
Hyperboli 
Voronoi diagram in Klein disk : 

 
2014 Frank NIELSEN 5793b870 24/57
Hyperboli 
Voronoi diagrams [19, 22℄ 
Ane Voronoi diagram equivalent to power diagram. 
Power distan 
e : kx − pk2 − wp (additively weighted ordinary 
Voronoi) 

 
2014 Frank NIELSEN 5793b870 25/57
Hyperboli 
Voronoi diagrams [19, 22℄ 
5 
ommon models of the abstra 
t hyperboli 
geometry 
https://www.youtube. 
om/wat 
h?v=i9IUzNxeH4o 
ACM Symposium on Computational Geometry (SoCG'14) 

 
2014 Frank NIELSEN 5793b870 26/57
k-order hyperboli 
Voronoi diagram 
k = 1 k = 2 
Ane diagram ⇒ equivalent to a power diagram [21℄ 
Wat 
h k-order Voronoi : 
https://www.youtube. 
om/wat 
h?v=i9IUzNxeH4o 

 
2014 Frank NIELSEN 5793b870 27/57
Constrast fun 
tion and dually at spa 
e 
◮ Convex and stri 
tly dierentiable fun 
tion F() admits a 
Legendre-Fen 
hel 
onvex 
onjugate F∗() : 
F∗() = sup 
 
(⊤ − F()), ∇F() =  = (∇F∗)−1() 
◮ Young's inequality gives rise to 
anoni 
al divergen 
e [8℄ : 
F()+F∗(′) ≥ ⊤′ ⇒ AF,F∗(, ′) = F() + F∗(′) − ⊤′ 
◮ Writing using single 
oordinate system, get dual Bregman 
divergen 
es : 
BF (p : q) = F(p) − F(q) − (p − q)⊤∇F(q) 
= BF∗(q : p) 
= AF,F∗(p, q) = AF∗,F (q : p) 

 
2014 Frank NIELSEN 5793b870 28/57
Dis 
riminant analysis exponential families 
◮ Consider X1 ∼ EF (1), X2 ∼ EF (2) (or moment 
parameterization i = ∇F(i )) 
◮ Bayes' rule : Classify x from X1 i. P(x|1)  P(x|2) 
◮ Use bije 
tion between exponential families and Bregman 
divergen 
es : 
log p(x|) = −BF∗(t(x) : ) + F∗(t(x)) + k(x) 
◮ MAP rule (w1 = w2 = 12 
) : 
BF∗(t(x) : 1)  BF∗(t(x) : 2) 
◮ Bregman bise 
tor : 
BiF∗(1, 2) = { |BF∗ ( : 1) = BF∗( : 1)} 

 
2014 Frank NIELSEN 5793b870 29/57
Bregman dual bise 
tors [3, 17, 20℄ 
Right-sided bise 
tor : → Hyperplane (-hyperplane) 
HF (p, q) = {x ∈ X | BF (x : p) = BF (x : q)}. 
HF : 
h∇F(p) − ∇F(q), xi + (F(p) − F(q) + hq,∇F(q)i − hp,∇F(p)i) = 0 
Left-sided bise 
tor : → -Hypersurfa 
e (-hyperplane) 
H′F(p, q) = {x ∈ X | BF (p : x) = BF (q : x)} 
H′F 
: h∇F(x), q − pi + F(p) − F(q) = 0 

 
2014 Frank NIELSEN 5793b870 30/57
Visualizing Bregman bise 
tors 
Primal 
oordinates  Dual 
oordinates  
natural parameters expe 
tation parameters 
p 
q 
Source Space: Itakura-Saito 
p(0.52977081,0.72041688) q(0.85824458,0.29083834) 
D(p,q)=0.66969016 D(q,p)=0.44835617 
Gradient Space: Itakura-Saito dual 
p’(-1.88760873,-1.38808518) q’(-1.16516903,-3.43833618) 
D*(p’,q’)=0.44835617 D*(q’,p’)=0.66969016 
p’ 
q’ 

 
2014 Frank NIELSEN 5793b870 31/57
Spa 
e of Bregman spheres and Bregman balls [3℄ 
Dual Bregman balls (bounding Bregman spheres) : 
Ballr 
F (c, r ) = {x ∈ X | BF (x : c) ≤ r} 
and Balll 
F (c, r ) = {x ∈ X | BF (c : x) ≤ r} 
Legendre duality : 
F (c, r ) = (∇F)−1(Ballr 
F∗(∇F(c), r )) 
Balll 
Illustration for Itakura-Saito divergen 
e, F(x) = −log x 

 
2014 Frank NIELSEN 5793b870 32/57
Lifting/Polarity : Potential fun 
tion graph F 

 
2014 Frank NIELSEN 5793b870 33/57
Spa 
e of Bregman spheres : Lifting map [3℄ 
F : x7→ ˆx = (x, F(x)), hypersurfa 
e in Rd+1. 
Hp : Tangent hyperplane at ˆp, z = Hp(x) = hx −p,∇F(p)i+F(p) 
◮ Bregman sphere  −→ ˆ with supporting hyperplane 
H : z = hx − c,∇F(c)i + F(c) + r . 
(// to Hc and shifted verti 
ally by r ) 
ˆ = F ∩ H. 
◮ interse 
tion of any hyperplane H with F proje 
ts onto X as a 
Bregman sphere : 
H : z = hx, ai+b →  : BallF (c = (∇F)−1(a), r = ha, ci−F(c)+b) 

 
2014 Frank NIELSEN 5793b870 34/57
Lifting with Itakura-Saito potential fun 
tion 

 
2014 Frank NIELSEN 5793b870 35/57
Spa 
e of Bregman spheres [3℄ 
◮ Union/interse 
tion of Bregman spheres from representational 
polytope [3℄ 
◮ Radi 
al axis of two Bregman balls is an hyperplane : 
Appli 
ations to Nearest Neighbor sear 
h trees like Bregman 
ball trees or Bregman vantage point trees [23℄. 

 
2014 Frank NIELSEN 5793b870 36/57
Spa 
e of spheres : Minimum en 
losing ball [14, 24℄ 
To a hyperplane H = H(a, b) : z = ha, xi+b in Rd+1, 
orresponds 
a ball  = Ball(c, r ) in Rd with 
enter c = ∇F∗(a) and radius : 
r = ha, ci−F(c)+b = ha,∇F∗(a)i−F(∇F∗(a))+b = F∗(a)+b 
sin 
e F(∇F∗(a)) = h∇F∗(a), ai − F∗(a) (Young equality) 
SEB : Find halfspa 
e H(a, b)− : z ≤ ha, xi + b that 
ontains all 
lifted points : 
min 
a,b 
r = F∗(a) + b, 
∀i ∈ {1, ..., n}, ha, xi i + b − F(xi ) ≥ 0. 
⇒ Convex Program (CP) with linear inequality 
onstraints 
⇒ F() = F∗() = 12 
x⊤x : CP → Quadrati 
Programming 
(QP) [4℄. 

 
2014 Frank NIELSEN 5793b870 37/57
InSphere predi 
ates wrt Bregman divergen 
es [3℄ 
Impli 
it representation of Bregman spheres/balls : 
◮ Is x inside the Bregman ball dened by d + 1 support 
points ? 
InSphere(x; p0, ..., pd ) =
1 ... 1 1 
p0 ... pd x 
F(p0) ... F(pd ) F(x)
◮ InSphere(x; p0, ..., pd ) is negative, null or positive depending 
on whether x lies inside, on, or outside . 

 
2014 Frank NIELSEN 5793b870 38/57
Bregman Voronoi diagrams as minimization diagrams [3℄ 
A sub 
lass of ane diagrams whi 
h have all non-empty 
ells . 
Minimization diagram of the n fun 
tions 
Di (x) = BF (x : pi ) = F(x) − F(pi ) − hx − pi ,∇F(pi )i. 
≡ minimization of n linear fun 
tions : 
Hi (x) = (pi − x)T∇F(qi ) − F(pi ) 
⇐⇒ 
Wat 
h https://www.youtube. 
om/wat 
h?v=L7v1wuN_9Wg 

 
2014 Frank NIELSEN 5793b870 39/57
Bregman Voronoi from Power diagrams [15, 20℄ 
Any ane diagram 
an be built from a power diagram. 
(power diagrams dened in full spa 
e Rd ) 
◮ Power distan 
e of x to Ball(p, r ) : ||p − x||2 − r 2. 
◮ Laguerre diagram : minimization diagram of 
Di (x) = ||pi − x||2 − r 2 
i 
◮ Power bise 
tor of Ball(pi , ri ) and Ball(pj , rj )= radi 
al 
hyperplane : 
2hx, pj − pi i + ||pi ||2 − ||pj ||2 + r 2 
j − r 2 
i = 0. 
Universality : Ane Bregman Voronoi diagram ⇔ Power diagram 

 
2014 Frank NIELSEN 5793b870 40/57
Ane Bregman Voronoi diagrams as power diagrams 
Equivalen 
e : B(∇F(pi ), ri ) with 
r 2 
i = h∇F(pi ),∇F(pi )i + 2(F(pi ) − hpi ,∇F (pi )i 
(imaginary radii shown in red, WLOG. ri ≥ 0 by shifting) 
Some 
ells may be empty in the Laguerre diagram (power diagram) 
but not in the Bregman diagram 
http://www. 
sl.sony. 
o.jp/person/nielsen/BVDapplet/ 

 
2014 Frank NIELSEN 5793b870 41/57
Bregman dual Delaunay triangulations 
Delaunay Exponential Hellinger-like 
◮ empty Bregman sphere property, 
◮ geodesi 
triangles. 
BVDs extends Eu 
lidean Voronoi diagrams with similar 
omplexity/algorithms. 

 
2014 Frank NIELSEN 5793b870 42/57
Riemannian tensor and orthogonality 
Dually at spa 
e. 
◮ F() 
onvex on  yields 
onvex 
onjugate F∗() on H 
◮ Same Riemannian tensor 
an be expressed in both 
oordinate 
systems : 
g() = ∇2F(), g∗() = ∇2F(), g((X))g∗((X)) = I , ∀X 
Same innitesimal length element ds2 = (ds∗)2. 
◮ Two 
urves 
1(t) and 
2(t) are orthogonal at 
Q = 
1(t0) = 
2(t0) i. 
h ˙ 
1(t0), ˙ 
2(t0)ig((Q)) = h ˙ 
1(t0), ˙ 
2(t0)i∗g 
∗((Q)) = 0 
◮ Geodesi 

(P,Q) and dual geodesi 

∗(P,Q) (and geodesi 
segments 
[P,Q] and 
∗[P,Q]) 

 
2014 Frank NIELSEN 5793b870 43/57
Orthogonality 
3-point property (generalized law of 
osines) : 
BF (p : r ) = BF (p : q) + BF (q : r ) − (p − q)T (∇F(r ) − ∇F(q))) 
P 
Q 
R 

[P,Q] 

[Q,R] 

(P,Q) orthogonal to 
∗(Q, R) i. 
BF (p : r ) = BF (p : q) + BF (q : r ) 
Equivalent to hp − q, r − qi = 0 
Extend Pythagoras' theorem 

(P,Q) ⊥ 
∗(Q, R) 

 
2014 Frank NIELSEN 5793b870 44/57
Dually orthogonal Bregman Voronoi  Triangulations 
Ordinary Voronoi diagram is perpendi 
ular to Delaunay 
triangulation. (Vor k-fa 
e ⊥ Del d − k-fa 
e) 
Dual line segment geodesi 
s : 

(P,Q) = { = p + (1 − )q | ∈ [0, 1]} 

∗(P,Q) = { = p + (1 − )q | ∈ [0, 1]} 
Bise 
tors : 
Bi(p, q) : hx, q − pi + F(p) − F(q) = 0 
Bi(p, q) : hx, q − pi + F∗(p) − F∗(q) = 0 
Dual orthogonality : 
Bi(p, q) ⊥ 
∗(P,Q) 

(P,Q) ⊥ Bi(p, q) 

 
2014 Frank NIELSEN 5793b870 45/57
Dually orthogonal Bregman Voronoi  Triangulations 
Bi(p, q) ⊥ 
∗(P,Q) 

(P,Q) ⊥ Bi(p, q) 

 
2014 Frank NIELSEN 5793b870 46/57
MAP rule and additive Bregman Voronoi diagrams 
Optimal MAP de 
ision rule : 
map(x) = argmaxi∈{1,...,n}wipi (x) 
= argmaxi∈{1,...,n} − B∗(t(x) : i ) + log wi , 
= argmini∈{1,...,n}B∗(t(x) : i ) − log wi 
⇒dual Bregman Voronoi with additive weights (ane 
diagram). 
Verti 
al proje 
tion of the interse 
tions of the lifted half-spa 
es on 
the potential F∗ shifted by the weights −log wi . 

 
2014 Frank NIELSEN 5793b870 47/57
Relative entropy for exponential families [18℄ 
◮ Kullba 
k-Leibler divergen 
e ( 
ross-entropy minus entropy) : 
KL(P : Q) = 
Z 
p(x) log 
p(x) 
q(x) 
dx ≥ 0 
= 
Z 
p(x) log 
1 
q(x) 
dx 
| {z } 
H×(P:Q)+c 
− 
Z 
p(x) log 
1 
p(x) 
dx 
| {z } 
H(p)=H×(P:P)−c 
= F(Q) − F(P) − hQ − P,∇F(P)i 
= BF (Q : P) = BF∗(P : Q) 
Bregman divergen 
e BF dened for a stri 
tly 
onvex and 
dierentiable fun 
tion up to some ane terms. 
◮ Proof KL(P : Q) = BF (Q : P) follows from 
X ∼ EF () =⇒ E[t(X)] = ∇F() =  

 
2014 Frank NIELSEN 5793b870 48/57

More Related Content

What's hot

asymptotics of ABC
asymptotics of ABCasymptotics of ABC
asymptotics of ABC
Christian Robert
 
Interpolation functions
Interpolation functionsInterpolation functions
Interpolation functions
Tarun Gehlot
 
CISEA 2019: ABC consistency and convergence
CISEA 2019: ABC consistency and convergenceCISEA 2019: ABC consistency and convergence
CISEA 2019: ABC consistency and convergence
Christian Robert
 
Monte Carlo in Montréal 2017
Monte Carlo in Montréal 2017Monte Carlo in Montréal 2017
Monte Carlo in Montréal 2017
Christian Robert
 
Inference in generative models using the Wasserstein distance [[INI]
Inference in generative models using the Wasserstein distance [[INI]Inference in generative models using the Wasserstein distance [[INI]
Inference in generative models using the Wasserstein distance [[INI]
Christian Robert
 
Statistics (1): estimation Chapter 3: likelihood function and likelihood esti...
Statistics (1): estimation Chapter 3: likelihood function and likelihood esti...Statistics (1): estimation Chapter 3: likelihood function and likelihood esti...
Statistics (1): estimation Chapter 3: likelihood function and likelihood esti...
Christian Robert
 
the ABC of ABC
the ABC of ABCthe ABC of ABC
the ABC of ABC
Christian Robert
 
ABC-Gibbs
ABC-GibbsABC-Gibbs
ABC-Gibbs
Christian Robert
 
Slides: Hypothesis testing, information divergence and computational geometry
Slides: Hypothesis testing, information divergence and computational geometrySlides: Hypothesis testing, information divergence and computational geometry
Slides: Hypothesis testing, information divergence and computational geometryFrank Nielsen
 
Applied numerical methods lec9
Applied numerical methods lec9Applied numerical methods lec9
Applied numerical methods lec9
Yasser Ahmed
 
Statistics (1): estimation, Chapter 1: Models
Statistics (1): estimation, Chapter 1: ModelsStatistics (1): estimation, Chapter 1: Models
Statistics (1): estimation, Chapter 1: Models
Christian Robert
 
ABC based on Wasserstein distances
ABC based on Wasserstein distancesABC based on Wasserstein distances
ABC based on Wasserstein distances
Christian Robert
 
Multiple estimators for Monte Carlo approximations
Multiple estimators for Monte Carlo approximationsMultiple estimators for Monte Carlo approximations
Multiple estimators for Monte Carlo approximations
Christian Robert
 
Laplace's Demon: seminar #1
Laplace's Demon: seminar #1Laplace's Demon: seminar #1
Laplace's Demon: seminar #1
Christian Robert
 
Slides: The dual Voronoi diagrams with respect to representational Bregman di...
Slides: The dual Voronoi diagrams with respect to representational Bregman di...Slides: The dual Voronoi diagrams with respect to representational Bregman di...
Slides: The dual Voronoi diagrams with respect to representational Bregman di...
Frank Nielsen
 
Newton divided difference interpolation
Newton divided difference interpolationNewton divided difference interpolation
Newton divided difference interpolation
VISHAL DONGA
 
ABC-Gibbs
ABC-GibbsABC-Gibbs
ABC-Gibbs
Christian Robert
 
Tailored Bregman Ball Trees for Effective Nearest Neighbors
Tailored Bregman Ball Trees for Effective Nearest NeighborsTailored Bregman Ball Trees for Effective Nearest Neighbors
Tailored Bregman Ball Trees for Effective Nearest Neighbors
Frank Nielsen
 
Can we estimate a constant?
Can we estimate a constant?Can we estimate a constant?
Can we estimate a constant?
Christian Robert
 
Interpolation In Numerical Methods.
 Interpolation In Numerical Methods. Interpolation In Numerical Methods.
Interpolation In Numerical Methods.
Abu Kaisar
 

What's hot (20)

asymptotics of ABC
asymptotics of ABCasymptotics of ABC
asymptotics of ABC
 
Interpolation functions
Interpolation functionsInterpolation functions
Interpolation functions
 
CISEA 2019: ABC consistency and convergence
CISEA 2019: ABC consistency and convergenceCISEA 2019: ABC consistency and convergence
CISEA 2019: ABC consistency and convergence
 
Monte Carlo in Montréal 2017
Monte Carlo in Montréal 2017Monte Carlo in Montréal 2017
Monte Carlo in Montréal 2017
 
Inference in generative models using the Wasserstein distance [[INI]
Inference in generative models using the Wasserstein distance [[INI]Inference in generative models using the Wasserstein distance [[INI]
Inference in generative models using the Wasserstein distance [[INI]
 
Statistics (1): estimation Chapter 3: likelihood function and likelihood esti...
Statistics (1): estimation Chapter 3: likelihood function and likelihood esti...Statistics (1): estimation Chapter 3: likelihood function and likelihood esti...
Statistics (1): estimation Chapter 3: likelihood function and likelihood esti...
 
the ABC of ABC
the ABC of ABCthe ABC of ABC
the ABC of ABC
 
ABC-Gibbs
ABC-GibbsABC-Gibbs
ABC-Gibbs
 
Slides: Hypothesis testing, information divergence and computational geometry
Slides: Hypothesis testing, information divergence and computational geometrySlides: Hypothesis testing, information divergence and computational geometry
Slides: Hypothesis testing, information divergence and computational geometry
 
Applied numerical methods lec9
Applied numerical methods lec9Applied numerical methods lec9
Applied numerical methods lec9
 
Statistics (1): estimation, Chapter 1: Models
Statistics (1): estimation, Chapter 1: ModelsStatistics (1): estimation, Chapter 1: Models
Statistics (1): estimation, Chapter 1: Models
 
ABC based on Wasserstein distances
ABC based on Wasserstein distancesABC based on Wasserstein distances
ABC based on Wasserstein distances
 
Multiple estimators for Monte Carlo approximations
Multiple estimators for Monte Carlo approximationsMultiple estimators for Monte Carlo approximations
Multiple estimators for Monte Carlo approximations
 
Laplace's Demon: seminar #1
Laplace's Demon: seminar #1Laplace's Demon: seminar #1
Laplace's Demon: seminar #1
 
Slides: The dual Voronoi diagrams with respect to representational Bregman di...
Slides: The dual Voronoi diagrams with respect to representational Bregman di...Slides: The dual Voronoi diagrams with respect to representational Bregman di...
Slides: The dual Voronoi diagrams with respect to representational Bregman di...
 
Newton divided difference interpolation
Newton divided difference interpolationNewton divided difference interpolation
Newton divided difference interpolation
 
ABC-Gibbs
ABC-GibbsABC-Gibbs
ABC-Gibbs
 
Tailored Bregman Ball Trees for Effective Nearest Neighbors
Tailored Bregman Ball Trees for Effective Nearest NeighborsTailored Bregman Ball Trees for Effective Nearest Neighbors
Tailored Bregman Ball Trees for Effective Nearest Neighbors
 
Can we estimate a constant?
Can we estimate a constant?Can we estimate a constant?
Can we estimate a constant?
 
Interpolation In Numerical Methods.
 Interpolation In Numerical Methods. Interpolation In Numerical Methods.
Interpolation In Numerical Methods.
 

Viewers also liked

Fundamentals cig 4thdec
Fundamentals cig 4thdecFundamentals cig 4thdec
Fundamentals cig 4thdec
Frank Nielsen
 
On Clustering Histograms with k-Means by Using Mixed α-Divergences
 On Clustering Histograms with k-Means by Using Mixed α-Divergences On Clustering Histograms with k-Means by Using Mixed α-Divergences
On Clustering Histograms with k-Means by Using Mixed α-Divergences
Frank Nielsen
 
(slides 9) Visual Computing: Geometry, Graphics, and Vision
(slides 9) Visual Computing: Geometry, Graphics, and Vision(slides 9) Visual Computing: Geometry, Graphics, and Vision
(slides 9) Visual Computing: Geometry, Graphics, and Vision
Frank Nielsen
 
On approximating the Riemannian 1-center
On approximating the Riemannian 1-centerOn approximating the Riemannian 1-center
On approximating the Riemannian 1-center
Frank Nielsen
 
(slides 8) Visual Computing: Geometry, Graphics, and Vision
(slides 8) Visual Computing: Geometry, Graphics, and Vision(slides 8) Visual Computing: Geometry, Graphics, and Vision
(slides 8) Visual Computing: Geometry, Graphics, and Vision
Frank Nielsen
 
k-MLE: A fast algorithm for learning statistical mixture models
k-MLE: A fast algorithm for learning statistical mixture modelsk-MLE: A fast algorithm for learning statistical mixture models
k-MLE: A fast algorithm for learning statistical mixture models
Frank Nielsen
 
On learning statistical mixtures maximizing the complete likelihood
On learning statistical mixtures maximizing the complete likelihoodOn learning statistical mixtures maximizing the complete likelihood
On learning statistical mixtures maximizing the complete likelihood
Frank Nielsen
 
A new implementation of k-MLE for mixture modelling of Wishart distributions
A new implementation of k-MLE for mixture modelling of Wishart distributionsA new implementation of k-MLE for mixture modelling of Wishart distributions
A new implementation of k-MLE for mixture modelling of Wishart distributions
Frank Nielsen
 
Computational Information Geometry: A quick review (ICMS)
Computational Information Geometry: A quick review (ICMS)Computational Information Geometry: A quick review (ICMS)
Computational Information Geometry: A quick review (ICMS)
Frank Nielsen
 
Traitement massif des données 2016
Traitement massif des données 2016Traitement massif des données 2016
Traitement massif des données 2016
Frank Nielsen
 
On representing spherical videos (Frank Nielsen, CVPR 2001)
On representing spherical videos (Frank Nielsen, CVPR 2001)On representing spherical videos (Frank Nielsen, CVPR 2001)
On representing spherical videos (Frank Nielsen, CVPR 2001)
Frank Nielsen
 
Divergence center-based clustering and their applications
Divergence center-based clustering and their applicationsDivergence center-based clustering and their applications
Divergence center-based clustering and their applications
Frank Nielsen
 
Classification with mixtures of curved Mahalanobis metrics
Classification with mixtures of curved Mahalanobis metricsClassification with mixtures of curved Mahalanobis metrics
Classification with mixtures of curved Mahalanobis metrics
Frank Nielsen
 
Traitement des données massives (INF442, A8)
Traitement des données massives (INF442, A8)Traitement des données massives (INF442, A8)
Traitement des données massives (INF442, A8)
Frank Nielsen
 
Divergence clustering
Divergence clusteringDivergence clustering
Divergence clustering
Frank Nielsen
 
Traitement des données massives (INF442, A5)
Traitement des données massives (INF442, A5)Traitement des données massives (INF442, A5)
Traitement des données massives (INF442, A5)
Frank Nielsen
 
The dual geometry of Shannon information
The dual geometry of Shannon informationThe dual geometry of Shannon information
The dual geometry of Shannon information
Frank Nielsen
 
Patch Matching with Polynomial Exponential Families and Projective Divergences
Patch Matching with Polynomial Exponential Families and Projective DivergencesPatch Matching with Polynomial Exponential Families and Projective Divergences
Patch Matching with Polynomial Exponential Families and Projective Divergences
Frank Nielsen
 
Traitement des données massives (INF442, A7)
Traitement des données massives (INF442, A7)Traitement des données massives (INF442, A7)
Traitement des données massives (INF442, A7)
Frank Nielsen
 
Traitement des données massives (INF442, A6)
Traitement des données massives (INF442, A6)Traitement des données massives (INF442, A6)
Traitement des données massives (INF442, A6)
Frank Nielsen
 

Viewers also liked (20)

Fundamentals cig 4thdec
Fundamentals cig 4thdecFundamentals cig 4thdec
Fundamentals cig 4thdec
 
On Clustering Histograms with k-Means by Using Mixed α-Divergences
 On Clustering Histograms with k-Means by Using Mixed α-Divergences On Clustering Histograms with k-Means by Using Mixed α-Divergences
On Clustering Histograms with k-Means by Using Mixed α-Divergences
 
(slides 9) Visual Computing: Geometry, Graphics, and Vision
(slides 9) Visual Computing: Geometry, Graphics, and Vision(slides 9) Visual Computing: Geometry, Graphics, and Vision
(slides 9) Visual Computing: Geometry, Graphics, and Vision
 
On approximating the Riemannian 1-center
On approximating the Riemannian 1-centerOn approximating the Riemannian 1-center
On approximating the Riemannian 1-center
 
(slides 8) Visual Computing: Geometry, Graphics, and Vision
(slides 8) Visual Computing: Geometry, Graphics, and Vision(slides 8) Visual Computing: Geometry, Graphics, and Vision
(slides 8) Visual Computing: Geometry, Graphics, and Vision
 
k-MLE: A fast algorithm for learning statistical mixture models
k-MLE: A fast algorithm for learning statistical mixture modelsk-MLE: A fast algorithm for learning statistical mixture models
k-MLE: A fast algorithm for learning statistical mixture models
 
On learning statistical mixtures maximizing the complete likelihood
On learning statistical mixtures maximizing the complete likelihoodOn learning statistical mixtures maximizing the complete likelihood
On learning statistical mixtures maximizing the complete likelihood
 
A new implementation of k-MLE for mixture modelling of Wishart distributions
A new implementation of k-MLE for mixture modelling of Wishart distributionsA new implementation of k-MLE for mixture modelling of Wishart distributions
A new implementation of k-MLE for mixture modelling of Wishart distributions
 
Computational Information Geometry: A quick review (ICMS)
Computational Information Geometry: A quick review (ICMS)Computational Information Geometry: A quick review (ICMS)
Computational Information Geometry: A quick review (ICMS)
 
Traitement massif des données 2016
Traitement massif des données 2016Traitement massif des données 2016
Traitement massif des données 2016
 
On representing spherical videos (Frank Nielsen, CVPR 2001)
On representing spherical videos (Frank Nielsen, CVPR 2001)On representing spherical videos (Frank Nielsen, CVPR 2001)
On representing spherical videos (Frank Nielsen, CVPR 2001)
 
Divergence center-based clustering and their applications
Divergence center-based clustering and their applicationsDivergence center-based clustering and their applications
Divergence center-based clustering and their applications
 
Classification with mixtures of curved Mahalanobis metrics
Classification with mixtures of curved Mahalanobis metricsClassification with mixtures of curved Mahalanobis metrics
Classification with mixtures of curved Mahalanobis metrics
 
Traitement des données massives (INF442, A8)
Traitement des données massives (INF442, A8)Traitement des données massives (INF442, A8)
Traitement des données massives (INF442, A8)
 
Divergence clustering
Divergence clusteringDivergence clustering
Divergence clustering
 
Traitement des données massives (INF442, A5)
Traitement des données massives (INF442, A5)Traitement des données massives (INF442, A5)
Traitement des données massives (INF442, A5)
 
The dual geometry of Shannon information
The dual geometry of Shannon informationThe dual geometry of Shannon information
The dual geometry of Shannon information
 
Patch Matching with Polynomial Exponential Families and Projective Divergences
Patch Matching with Polynomial Exponential Families and Projective DivergencesPatch Matching with Polynomial Exponential Families and Projective Divergences
Patch Matching with Polynomial Exponential Families and Projective Divergences
 
Traitement des données massives (INF442, A7)
Traitement des données massives (INF442, A7)Traitement des données massives (INF442, A7)
Traitement des données massives (INF442, A7)
 
Traitement des données massives (INF442, A6)
Traitement des données massives (INF442, A6)Traitement des données massives (INF442, A6)
Traitement des données massives (INF442, A6)
 

Similar to Voronoi diagrams in information geometry:  Statistical Voronoi diagrams and their applications

Multivriada ppt ms
Multivriada   ppt msMultivriada   ppt ms
Multivriada ppt ms
Faeco Bot
 
Slides: A glance at information-geometric signal processing
Slides: A glance at information-geometric signal processingSlides: A glance at information-geometric signal processing
Slides: A glance at information-geometric signal processing
Frank Nielsen
 
Testing for mixtures by seeking components
Testing for mixtures by seeking componentsTesting for mixtures by seeking components
Testing for mixtures by seeking components
Christian Robert
 
Slides: On the Chi Square and Higher-Order Chi Distances for Approximating f-...
Slides: On the Chi Square and Higher-Order Chi Distances for Approximating f-...Slides: On the Chi Square and Higher-Order Chi Distances for Approximating f-...
Slides: On the Chi Square and Higher-Order Chi Distances for Approximating f-...
Frank Nielsen
 
ABC with Wasserstein distances
ABC with Wasserstein distancesABC with Wasserstein distances
ABC with Wasserstein distances
Christian Robert
 
Slides: Total Jensen divergences: Definition, Properties and k-Means++ Cluste...
Slides: Total Jensen divergences: Definition, Properties and k-Means++ Cluste...Slides: Total Jensen divergences: Definition, Properties and k-Means++ Cluste...
Slides: Total Jensen divergences: Definition, Properties and k-Means++ Cluste...
Frank Nielsen
 
Basics of probability in statistical simulation and stochastic programming
Basics of probability in statistical simulation and stochastic programmingBasics of probability in statistical simulation and stochastic programming
Basics of probability in statistical simulation and stochastic programming
SSA KPI
 
QMC: Operator Splitting Workshop, Stochastic Block-Coordinate Fixed Point Alg...
QMC: Operator Splitting Workshop, Stochastic Block-Coordinate Fixed Point Alg...QMC: Operator Splitting Workshop, Stochastic Block-Coordinate Fixed Point Alg...
QMC: Operator Splitting Workshop, Stochastic Block-Coordinate Fixed Point Alg...
The Statistical and Applied Mathematical Sciences Institute
 
Quantum chaos of generic systems - Marko Robnik
Quantum chaos of generic systems - Marko RobnikQuantum chaos of generic systems - Marko Robnik
Quantum chaos of generic systems - Marko Robnik
Lake Como School of Advanced Studies
 
IVR - Chapter 1 - Introduction
IVR - Chapter 1 - IntroductionIVR - Chapter 1 - Introduction
IVR - Chapter 1 - Introduction
Charles Deledalle
 
BAYSM'14, Wien, Austria
BAYSM'14, Wien, AustriaBAYSM'14, Wien, Austria
BAYSM'14, Wien, Austria
Christian Robert
 
A sharp nonlinear Hausdorff-Young inequality for small potentials
A sharp nonlinear Hausdorff-Young inequality for small potentialsA sharp nonlinear Hausdorff-Young inequality for small potentials
A sharp nonlinear Hausdorff-Young inequality for small potentials
VjekoslavKovac1
 
Litvinenko_RWTH_UQ_Seminar_talk.pdf
Litvinenko_RWTH_UQ_Seminar_talk.pdfLitvinenko_RWTH_UQ_Seminar_talk.pdf
Litvinenko_RWTH_UQ_Seminar_talk.pdf
Alexander Litvinenko
 
QMC: Operator Splitting Workshop, Compactness Estimates for Nonlinear PDEs - ...
QMC: Operator Splitting Workshop, Compactness Estimates for Nonlinear PDEs - ...QMC: Operator Splitting Workshop, Compactness Estimates for Nonlinear PDEs - ...
QMC: Operator Splitting Workshop, Compactness Estimates for Nonlinear PDEs - ...
The Statistical and Applied Mathematical Sciences Institute
 
Distributions
DistributionsDistributions
Distributions
adnan gilani
 
Optimal interval clustering: Application to Bregman clustering and statistica...
Optimal interval clustering: Application to Bregman clustering and statistica...Optimal interval clustering: Application to Bregman clustering and statistica...
Optimal interval clustering: Application to Bregman clustering and statistica...
Frank Nielsen
 
Finance Enginering from Columbia.pdf
Finance Enginering from Columbia.pdfFinance Enginering from Columbia.pdf
Finance Enginering from Columbia.pdf
CarlosLazo45
 
Density theorems for anisotropic point configurations
Density theorems for anisotropic point configurationsDensity theorems for anisotropic point configurations
Density theorems for anisotropic point configurations
VjekoslavKovac1
 
Approximation Methods Of Solutions For Equilibrium Problem In Hilbert Spaces
Approximation Methods Of Solutions For Equilibrium Problem In Hilbert SpacesApproximation Methods Of Solutions For Equilibrium Problem In Hilbert Spaces
Approximation Methods Of Solutions For Equilibrium Problem In Hilbert Spaces
Lisa Garcia
 

Similar to Voronoi diagrams in information geometry:  Statistical Voronoi diagrams and their applications (20)

Multivriada ppt ms
Multivriada   ppt msMultivriada   ppt ms
Multivriada ppt ms
 
Slides: A glance at information-geometric signal processing
Slides: A glance at information-geometric signal processingSlides: A glance at information-geometric signal processing
Slides: A glance at information-geometric signal processing
 
Testing for mixtures by seeking components
Testing for mixtures by seeking componentsTesting for mixtures by seeking components
Testing for mixtures by seeking components
 
Slides: On the Chi Square and Higher-Order Chi Distances for Approximating f-...
Slides: On the Chi Square and Higher-Order Chi Distances for Approximating f-...Slides: On the Chi Square and Higher-Order Chi Distances for Approximating f-...
Slides: On the Chi Square and Higher-Order Chi Distances for Approximating f-...
 
ABC with Wasserstein distances
ABC with Wasserstein distancesABC with Wasserstein distances
ABC with Wasserstein distances
 
Slides: Total Jensen divergences: Definition, Properties and k-Means++ Cluste...
Slides: Total Jensen divergences: Definition, Properties and k-Means++ Cluste...Slides: Total Jensen divergences: Definition, Properties and k-Means++ Cluste...
Slides: Total Jensen divergences: Definition, Properties and k-Means++ Cluste...
 
Puy chosuantai2
Puy chosuantai2Puy chosuantai2
Puy chosuantai2
 
Basics of probability in statistical simulation and stochastic programming
Basics of probability in statistical simulation and stochastic programmingBasics of probability in statistical simulation and stochastic programming
Basics of probability in statistical simulation and stochastic programming
 
QMC: Operator Splitting Workshop, Stochastic Block-Coordinate Fixed Point Alg...
QMC: Operator Splitting Workshop, Stochastic Block-Coordinate Fixed Point Alg...QMC: Operator Splitting Workshop, Stochastic Block-Coordinate Fixed Point Alg...
QMC: Operator Splitting Workshop, Stochastic Block-Coordinate Fixed Point Alg...
 
Quantum chaos of generic systems - Marko Robnik
Quantum chaos of generic systems - Marko RobnikQuantum chaos of generic systems - Marko Robnik
Quantum chaos of generic systems - Marko Robnik
 
IVR - Chapter 1 - Introduction
IVR - Chapter 1 - IntroductionIVR - Chapter 1 - Introduction
IVR - Chapter 1 - Introduction
 
BAYSM'14, Wien, Austria
BAYSM'14, Wien, AustriaBAYSM'14, Wien, Austria
BAYSM'14, Wien, Austria
 
A sharp nonlinear Hausdorff-Young inequality for small potentials
A sharp nonlinear Hausdorff-Young inequality for small potentialsA sharp nonlinear Hausdorff-Young inequality for small potentials
A sharp nonlinear Hausdorff-Young inequality for small potentials
 
Litvinenko_RWTH_UQ_Seminar_talk.pdf
Litvinenko_RWTH_UQ_Seminar_talk.pdfLitvinenko_RWTH_UQ_Seminar_talk.pdf
Litvinenko_RWTH_UQ_Seminar_talk.pdf
 
QMC: Operator Splitting Workshop, Compactness Estimates for Nonlinear PDEs - ...
QMC: Operator Splitting Workshop, Compactness Estimates for Nonlinear PDEs - ...QMC: Operator Splitting Workshop, Compactness Estimates for Nonlinear PDEs - ...
QMC: Operator Splitting Workshop, Compactness Estimates for Nonlinear PDEs - ...
 
Distributions
DistributionsDistributions
Distributions
 
Optimal interval clustering: Application to Bregman clustering and statistica...
Optimal interval clustering: Application to Bregman clustering and statistica...Optimal interval clustering: Application to Bregman clustering and statistica...
Optimal interval clustering: Application to Bregman clustering and statistica...
 
Finance Enginering from Columbia.pdf
Finance Enginering from Columbia.pdfFinance Enginering from Columbia.pdf
Finance Enginering from Columbia.pdf
 
Density theorems for anisotropic point configurations
Density theorems for anisotropic point configurationsDensity theorems for anisotropic point configurations
Density theorems for anisotropic point configurations
 
Approximation Methods Of Solutions For Equilibrium Problem In Hilbert Spaces
Approximation Methods Of Solutions For Equilibrium Problem In Hilbert SpacesApproximation Methods Of Solutions For Equilibrium Problem In Hilbert Spaces
Approximation Methods Of Solutions For Equilibrium Problem In Hilbert Spaces
 

Recently uploaded

Hunza Cherry Blossom tour 2025- Hunza Adventure Tours
Hunza Cherry Blossom tour 2025- Hunza Adventure ToursHunza Cherry Blossom tour 2025- Hunza Adventure Tours
Hunza Cherry Blossom tour 2025- Hunza Adventure Tours
Hunza Adventure Tours
 
4 DAYS MASAI MARA WILDEBEEST MIGRATION SAFARI TOUR PACKAGE KENYA
4 DAYS MASAI MARA WILDEBEEST MIGRATION SAFARI TOUR PACKAGE KENYA4 DAYS MASAI MARA WILDEBEEST MIGRATION SAFARI TOUR PACKAGE KENYA
4 DAYS MASAI MARA WILDEBEEST MIGRATION SAFARI TOUR PACKAGE KENYA
Bush Troop Safari
 
Exploring Heritage The Ultimate Cultural Tour in Palmer, Puerto Rico
Exploring Heritage The Ultimate Cultural Tour in Palmer, Puerto RicoExploring Heritage The Ultimate Cultural Tour in Palmer, Puerto Rico
Exploring Heritage The Ultimate Cultural Tour in Palmer, Puerto Rico
Caribbean Breeze Adventures
 
Winter Festivities in Italy
Winter Festivities in ItalyWinter Festivities in Italy
Winter Festivities in Italy
Time for Sicily
 
Exploring Montreal's Artistic Heritage Top Art Galleries and Museums to Visit
Exploring Montreal's Artistic Heritage Top Art Galleries and Museums to VisitExploring Montreal's Artistic Heritage Top Art Galleries and Museums to Visit
Exploring Montreal's Artistic Heritage Top Art Galleries and Museums to Visit
Spade & Palacio Tours
 
欧洲杯开户-信誉的欧洲杯开户-正规欧洲杯开户|【​网址​🎉ac123.net🎉​】
欧洲杯开户-信誉的欧洲杯开户-正规欧洲杯开户|【​网址​🎉ac123.net🎉​】欧洲杯开户-信誉的欧洲杯开户-正规欧洲杯开户|【​网址​🎉ac123.net🎉​】
欧洲杯开户-信誉的欧洲杯开户-正规欧洲杯开户|【​网址​🎉ac123.net🎉​】
bljeremy734
 
Paddle, Float, and Explore The Ultimate River Tour Experience in Monitor, WA
Paddle, Float, and Explore The Ultimate River Tour Experience in Monitor, WAPaddle, Float, and Explore The Ultimate River Tour Experience in Monitor, WA
Paddle, Float, and Explore The Ultimate River Tour Experience in Monitor, WA
River Recreation - Washington Whitewater Rafting
 
TOP 10 Historic Places To See in Kuruskhetra.
TOP 10 Historic Places To See in Kuruskhetra.TOP 10 Historic Places To See in Kuruskhetra.
TOP 10 Historic Places To See in Kuruskhetra.
ujjwalsethi113
 
Understanding the Running Costs of Electric Scooters.pptx
Understanding the Running Costs of Electric Scooters.pptxUnderstanding the Running Costs of Electric Scooters.pptx
Understanding the Running Costs of Electric Scooters.pptx
Zivah ElectriVa Private Limited
 
Antarctica- Icy wilderness of extremes and wonder
Antarctica- Icy wilderness of extremes and wonderAntarctica- Icy wilderness of extremes and wonder
Antarctica- Icy wilderness of extremes and wonder
tahreemzahra82
 
Agence Régionale du Tourisme Grand Est - brochure MICE 2024.pdf
Agence Régionale du Tourisme Grand Est - brochure MICE 2024.pdfAgence Régionale du Tourisme Grand Est - brochure MICE 2024.pdf
Agence Régionale du Tourisme Grand Est - brochure MICE 2024.pdf
MICEboard
 
MC INTERNATIONALS | TRAVEL COMPANY IN JHANG
MC INTERNATIONALS | TRAVEL COMPANY IN JHANGMC INTERNATIONALS | TRAVEL COMPANY IN JHANG
MC INTERNATIONALS | TRAVEL COMPANY IN JHANG
AshBhatt4
 
Jose RIZAL History and his travel Paris to berlin
Jose RIZAL History and his travel Paris to berlinJose RIZAL History and his travel Paris to berlin
Jose RIZAL History and his travel Paris to berlin
MaryGraceArdalesLope
 
TRAVEL TO MT. RWENZORI NATIONAL PARK WITH NILE ABENTEUER SAFARIS.docx
TRAVEL TO MT. RWENZORI NATIONAL PARK WITH NILE ABENTEUER SAFARIS.docxTRAVEL TO MT. RWENZORI NATIONAL PARK WITH NILE ABENTEUER SAFARIS.docx
TRAVEL TO MT. RWENZORI NATIONAL PARK WITH NILE ABENTEUER SAFARIS.docx
nileabenteuersafaris
 
How To Change Name On Volaris Ticket.pdf
How To Change Name On Volaris Ticket.pdfHow To Change Name On Volaris Ticket.pdf
How To Change Name On Volaris Ticket.pdf
namechange763
 
Get tailored experience with Stonehenge tours from London
Get tailored experience with Stonehenge tours from LondonGet tailored experience with Stonehenge tours from London
Get tailored experience with Stonehenge tours from London
London Country Tours
 

Recently uploaded (16)

Hunza Cherry Blossom tour 2025- Hunza Adventure Tours
Hunza Cherry Blossom tour 2025- Hunza Adventure ToursHunza Cherry Blossom tour 2025- Hunza Adventure Tours
Hunza Cherry Blossom tour 2025- Hunza Adventure Tours
 
4 DAYS MASAI MARA WILDEBEEST MIGRATION SAFARI TOUR PACKAGE KENYA
4 DAYS MASAI MARA WILDEBEEST MIGRATION SAFARI TOUR PACKAGE KENYA4 DAYS MASAI MARA WILDEBEEST MIGRATION SAFARI TOUR PACKAGE KENYA
4 DAYS MASAI MARA WILDEBEEST MIGRATION SAFARI TOUR PACKAGE KENYA
 
Exploring Heritage The Ultimate Cultural Tour in Palmer, Puerto Rico
Exploring Heritage The Ultimate Cultural Tour in Palmer, Puerto RicoExploring Heritage The Ultimate Cultural Tour in Palmer, Puerto Rico
Exploring Heritage The Ultimate Cultural Tour in Palmer, Puerto Rico
 
Winter Festivities in Italy
Winter Festivities in ItalyWinter Festivities in Italy
Winter Festivities in Italy
 
Exploring Montreal's Artistic Heritage Top Art Galleries and Museums to Visit
Exploring Montreal's Artistic Heritage Top Art Galleries and Museums to VisitExploring Montreal's Artistic Heritage Top Art Galleries and Museums to Visit
Exploring Montreal's Artistic Heritage Top Art Galleries and Museums to Visit
 
欧洲杯开户-信誉的欧洲杯开户-正规欧洲杯开户|【​网址​🎉ac123.net🎉​】
欧洲杯开户-信誉的欧洲杯开户-正规欧洲杯开户|【​网址​🎉ac123.net🎉​】欧洲杯开户-信誉的欧洲杯开户-正规欧洲杯开户|【​网址​🎉ac123.net🎉​】
欧洲杯开户-信誉的欧洲杯开户-正规欧洲杯开户|【​网址​🎉ac123.net🎉​】
 
Paddle, Float, and Explore The Ultimate River Tour Experience in Monitor, WA
Paddle, Float, and Explore The Ultimate River Tour Experience in Monitor, WAPaddle, Float, and Explore The Ultimate River Tour Experience in Monitor, WA
Paddle, Float, and Explore The Ultimate River Tour Experience in Monitor, WA
 
TOP 10 Historic Places To See in Kuruskhetra.
TOP 10 Historic Places To See in Kuruskhetra.TOP 10 Historic Places To See in Kuruskhetra.
TOP 10 Historic Places To See in Kuruskhetra.
 
Understanding the Running Costs of Electric Scooters.pptx
Understanding the Running Costs of Electric Scooters.pptxUnderstanding the Running Costs of Electric Scooters.pptx
Understanding the Running Costs of Electric Scooters.pptx
 
Antarctica- Icy wilderness of extremes and wonder
Antarctica- Icy wilderness of extremes and wonderAntarctica- Icy wilderness of extremes and wonder
Antarctica- Icy wilderness of extremes and wonder
 
Agence Régionale du Tourisme Grand Est - brochure MICE 2024.pdf
Agence Régionale du Tourisme Grand Est - brochure MICE 2024.pdfAgence Régionale du Tourisme Grand Est - brochure MICE 2024.pdf
Agence Régionale du Tourisme Grand Est - brochure MICE 2024.pdf
 
MC INTERNATIONALS | TRAVEL COMPANY IN JHANG
MC INTERNATIONALS | TRAVEL COMPANY IN JHANGMC INTERNATIONALS | TRAVEL COMPANY IN JHANG
MC INTERNATIONALS | TRAVEL COMPANY IN JHANG
 
Jose RIZAL History and his travel Paris to berlin
Jose RIZAL History and his travel Paris to berlinJose RIZAL History and his travel Paris to berlin
Jose RIZAL History and his travel Paris to berlin
 
TRAVEL TO MT. RWENZORI NATIONAL PARK WITH NILE ABENTEUER SAFARIS.docx
TRAVEL TO MT. RWENZORI NATIONAL PARK WITH NILE ABENTEUER SAFARIS.docxTRAVEL TO MT. RWENZORI NATIONAL PARK WITH NILE ABENTEUER SAFARIS.docx
TRAVEL TO MT. RWENZORI NATIONAL PARK WITH NILE ABENTEUER SAFARIS.docx
 
How To Change Name On Volaris Ticket.pdf
How To Change Name On Volaris Ticket.pdfHow To Change Name On Volaris Ticket.pdf
How To Change Name On Volaris Ticket.pdf
 
Get tailored experience with Stonehenge tours from London
Get tailored experience with Stonehenge tours from LondonGet tailored experience with Stonehenge tours from London
Get tailored experience with Stonehenge tours from London
 

Voronoi diagrams in information geometry:  Statistical Voronoi diagrams and their applications

  • 1. Voronoi diagrams in information geometry Statisti al Voronoi diagrams and their appli ations Frank NIELSEN É ole Polyte hnique Sony CSL e-mail: Frank.Nielsen a m.org MaxEnt 2014 2014 Frank NIELSEN 5793b870 1/57
  • 2. Outline of the tutorial 1. Eu lidean Voronoi diagrams 2. Dis riminant analysis, Mahalanobis Voronoi diagrams, and anisotropi Voronoi diagrams 3. Fisher-Hotelling-Rao Voronoi diagrams (Riemannian urved geometries) 4. Kullba k-Leibler Voronoi diagrams (Bregman Voronoi diagrams, dually at geometries) 5. Bayes' error, Cherno information and statisti al Voronoi diagrams 2014 Frank NIELSEN 5793b870 2/57
  • 3. Eu lidean (ordinary) Voronoi diagrams P = {P1, ..., Pn} : n distin t point generators in Eu lidean spa e Ed =1Bi+(Pi , Pj ) V (Pi ) = {X : DE (Pi ,X) ≤ DE (Pj ,X), ∀j6= i} = ∩ni DE (P,Q) = k(P) − (Q)k2 = qPd i=1(i (P) − i (Q))2 (P) = p : Cartesian oordinate system with j (Pi ) = p(j) i . Bise tors Bi(P,Q) = {X : DE (P,X) ≤ DE (Q,X)} : hyperplanes Voronoi diagram = ell omplex V (Pi )'s with their fa es ⇒ Many appli ations : rystal growth, odebook/quantization, mole ule interfa es/do king, motion planning, et . 2014 Frank NIELSEN 5793b870 3/57
  • 4. Voronoi diagrams and dual Delaunay simpli ial omplex ⇒ Empty sphere property, max min angle triangulation, et ⇒ General position : no (d + 2) points ospheri al ⇒ Voronoi dual Delaunay triangulation ⇒ Bise tor Bi(P,Q) perpendi ular ⊥ to segment [PQ] 2014 Frank NIELSEN 5793b870 4/57
  • 5. Minimum spanning tree ⊂ Delaunay triangulation All edges of the Eu lidean MST are Delaunay edges → Prim's greedy algorithm in Delaunay graph in O(n log n) Convex hull : boundary @ of the Delaunay triangulation 2014 Frank NIELSEN 5793b870 5/57
  • 6. Order k Voronoi diagrams All subsets of size k, Pk = P k = {K1, ...,KN} with N = n k . partition the spa e into non-empty k-order Voronoi ells : Vork (Ki ) = {x : ∀q ∈ Ki , ∀r ∈ PKi , D(x, q) ≤ D(x, r )} Combinatorial omplexity not yet settled ! ! ! (k-sets/levels) ≡ proje tion of k-levels of arrangements of hyperplanes in Rd+1. 2014 Frank NIELSEN 5793b870 6/57
  • 7. Farthest order Voronoi and Minimum En losing Ball Cir um enter/minmax enter on the farthest Voronoi diagram. Non-dierentiable optim. at the farthest Voronoi [16℄ 2014 Frank NIELSEN 5793b870 7/57
  • 8. Voronoi Delaunay : Complexity and algorithms ◮ Complexity : (n⌈ d 2 ⌉) (quadrati in 3D) Moment urve : t7→ (t, t2, .., td ), et . ◮ Constru tion : (n log n + n⌈ d 2 ⌉), output-sensitive algorithms (n log n + f ), not yet optimal output-sensitive algorithms. ◮ Voronoi diagram : VorD(P) = Vorf (D(·,·))(P) for a stri tly monotoni ally in reasing fun tion f (·) Vor. Eu lidean ≡ Vor. squared Eu lidean ⊂ Vor. Bregman ◮ Geometri toolbox : spa e of spheres (polarity, orthogonality between spheres, et .) 2014 Frank NIELSEN 5793b870 8/57
  • 9. Multiple Hypothesis Testing [9, 10℄ Given a random variable X with n hypothesis H1 : X ∼ P1 : ... Hn : X ∼ Pn de ide for a IID sample x1, ..., xm ∼ X whi h hypothesis holds true ? Pm orre t = 1 − Pm error Asymptoti regime : Error exponent lim m→∞ − 1 m log Pm e = 2014 Frank NIELSEN 5793b870 9/57
  • 10. Bayesian hypothesis testing ◮ Prior probabilities : wi = P(X ∼ Pi ) 0 (with Pn i=1 wi = 1) ◮ Conditional probabilities : P(X = x|X ∼ Pi ). P(X = x) = Xn i=1 P(X ∼ Pi )P(X = x|X ∼ Pi ) = Xn i=1 wiP(X|Pi ) ◮ Cost design matrix [cij ]= with ci ,j = ost of de iding Hi when in fa t Hj is true ◮ pi ,j (u(x)) = probability of making this de ision using rule u. 2014 Frank NIELSEN 5793b870 10/57
  • 11. Bayesian dete tor : MAP rule ◮ Minimize the expe ted ost : EX [c(r (x))], c(r (x)) = X i  wi X j6=i   ci ,jpi ,j (r (x)) ◮ Spe ial ase : Probability of error Pe : ci ,i = 0 and ci ,j = 1 for i6= j : Pe = EX   X i  wi X j6=i   pi ,j (r (x))   ◮ Maximum a posteriori probability (MAP) rule : map(x) = argmaxi∈{1,...,n} wipi (x) where pi (x) = P(X = x|X ∼ Pi ) are the onditional probabilities. → MAP Bayesian dete tor minimizes Pe over all rules [7℄ 2014 Frank NIELSEN 5793b870 11/57
  • 12. MVNs, MAP rule, and Mahalanobis distan e (1930) MultiVariate Normals (MVNs) ◮ Say, X1 ∼ N(μ1,) and X2 ∼ N(μ2,) ◮ Probability of error (mis lassi ation) for w1 = w2 = 12 : Pe = 1 2 (1 − TV(P1, P2)) = − 1 2 (P1, P2) (·) : standard normal distribution fun tion ◮ Mahalanobis metri distan e (·, ·) : 2(X1,X2) = (μ1 − μ2)⊤−1(μ1 − μ2) = μ⊤−1μ ◮ Measure the density overlapping (the greater, the lesser) ◮ generalize Eu lidean distan e ( = I ) ◮ 2 : only symmetri Bregman divergen e. ⇒ easy to approximate P˜e experimentally 2014 Frank NIELSEN 5793b870 12/57
  • 13. Mahalanobis Voronoi diagrams MHT for isotropi MVNs with uniform weight wi = 1 n Cholesky de omposition = LL⊤ 2(X1,X2) = (μ1 − μ2)⊤−1(μ1 − μ2), = (μ1 − μ2)⊤(L−1)⊤L−1(μ1 − μ1), = (L−1μ1 − L−1μ2)⊤(L−1μ1 − L−1μ2) 2(X1,X2) = DE (L−1μ1, L−1μ2) Mahalanobis Voronoi diagram : ◮ Cholesky de omposition : = LL⊤ ◮ Map P = {p1, ..., pn} to P′ = {p′ 1 , ..., p′n i = L−1x. } with p′ ◮ Ordinary Voronoi diagram VorE (P′) ◮ Map VorE (P′) to Vor(P) = LVorE (P′). 2014 Frank NIELSEN 5793b870 13/57
  • 14. Mahalanobis Voronoi diagrams a ount for both orrelation and dimension (feature) s aling Dual stru ture ≡ anisotropi Delaunay triangulation ⇒ empty ir umellipse property Re all : MAP rule from Voronoi diagram 2014 Frank NIELSEN 5793b870 14/57
  • 15. Anisotropi Voronoi diagrams [6℄ (MHT for MVNs) Classi ation : Generator Xi ∼ N(μi ,i ), wi = 1 n . Dis riminant fun tions are quadrati bise tors Orphans, islands, dual not a triangulation (need wedged/visibility onditions) Appli ations : Crystal growth in elds (not Riemannian smooth metri ) 2014 Frank NIELSEN 5793b870 15/57
  • 16. Statisti s : Estimators Given x1, ..., xn IID observations (from a population), estimate the underlying distribution X ∼ p. ◮ empiri al CDF : ˆ Fn(x) = 1 n P i I(−∞,x)(xi ). Glivenko-Cantelli theorem : supx | ˆ Fn(x) − F(x)| → 0 a.s. ◮ Fisher approa h : Density p belongs to a parametri family p(x|), D = #parameters (order) ◮ Method of moments : Mat h distribution moments with sample moments : EX [Xl ] = 1 n Xn i=1 xl i , ⇒ any D independent equations yields an estimator ◮ Fisher Maximum (log-)Likelihood Estimator MLE : max l (; x1, ..., xn) = max Yn i=1 p(xi ; ) 2014 Frank NIELSEN 5793b870 16/57
  • 17. Estimation : Varian e Lower Bound Whi h estimator ˆ n shall we hoose ? Estimator is also a random variable on a random ve tor ◮ Consistent : limn→∞ ˆn → ◮ Unbiased : E[ˆn] − = 0 ◮ Minimum square error (MSE) : E[(ˆ − )2] = B[ˆ]2 + V[ˆ] ◮ Fisher information matrix : I () = Ii ,j () = E @ @i log p(x|) @ @j log p(x|) ◮ Fré het Darmois Cramér-Rao lower bound for unbiased estimators : V[ ˆ n] 1 n I−1() ◮ E ient : estimator rea hes the Cramèr-Rao lower bound 2014 Frank NIELSEN 5793b870 17/57
  • 18. Su ient statisti s and exponential families ◮ Su ient statisti t(x) : P(x|t(x) = t, ) = P(x|t(x) = t) All information for inferen e ontained in t(x). (6= an illary) ◮ For univariate Gaussians, D = 2 : t1(x) = x and t2(x) = x2. ˆμ = ¯x = 1 n X i xi , ˆ2 = 1 n X i (xi −¯x)2 = 1 n X i x2 i ! −(¯x)2 ◮ Exponential families have nite dimensional su ient statisti s (Koopman) : Redu e n data to D statisti s. ∀x ∈ X, P(x|) = exp(⊤t(x) − F() + k(x)) F(·) : log-normalizer/ umulant/partition fun tion k(x) : auxiliary term for arrier measure 2014 Frank NIELSEN 5793b870 18/57
  • 19. Exponential families : Fisher information and MLE Common distributions are exponential families : Poisson, Gaussians, Gamma, Beta, Diri hlet, et . ◮ Fisher information : I () = ∇2F() ◮ MLE for an exponential family : ∇F() = 1 n X i t(Xi ) = MLE exists i. = ∇F() ∈ int(CH(X)) 2014 Frank NIELSEN 5793b870 19/57
  • 20. Population spa e : Hotelling (1930) [5℄ Rao (1945) [25℄ Birth of dierential-geometri methods in statisti s. ◮ Fisher information matrix (positive denite) an be used as a (smooth) Riemannian metri tensor. ◮ Distan e between two populations indexed by 1 and 2 : Riemannian distan e (metri length) ◮ Fisher-Hotelling-Rao (FHR) geodesi distan e used in lassi ation : Find the losest population to a given population ◮ Used in tests of signi an e (null versus alternative hypothesis), power of a test : P(reje t H0|H0 is false ) (surfa es in population spa es) 2014 Frank NIELSEN 5793b870 20/57
  • 21. Rao's distan e (1945, introdu ed by Hotelling 1930 [5℄) ◮ Innitesimal length element : ds2 = X ij gij ()didj = dT I ()d ◮ Geodesi and distan e are hard to expli itly al ulate : (p(x; 1), p(x; 2)) = min (s) (0)=1 (1)=2 Z 1 0 s d ds T I () d ds ds ◮ Metri property of , many tools [1℄ : Riemannian Log/Exp tangent/manifold mapping 2014 Frank NIELSEN 5793b870 21/57
  • 22. Fisher Rao Hotelling Voronoi : Riemannian Voronoi diagrams ◮ Lo ation-s ale 2D families have non-positive urvature (Hotelling, 1930) : FHR Voronoi diagrams amount to hyperboli Voronoi diagrams or Eu lidean diagrams (lo ation families only like isotropi Gaussians) ◮ Multinomial family has spheri al geometry on the positive orthant : Spheri al Voronoi diagram (stereographi proje tion ∝ Eu lidean Voronoi diagrams) ◮ Arbitrary families p(x|) : Geodesi s not in losed forms → limited omputational framework in pra ti e... 2014 Frank NIELSEN 5793b870 22/57
  • 23. Hyperboli Voronoi diagrams [19, 22℄ ◮ In Klein disk, the hyperboli Voronoi diagram amounts to a lipped ane Voronoi diagram, or a lipped power diagram. E ient lipping algorithm [2℄. ◮ Convert to other models of hyperboli geometry : Poin aré disk, upper half spa e, hyperboloid, Beltrami hemisphere, et . ◮ Conformal (good for vizualizing) versus non- onformal (good for omputing) models. ( onformal metri G(x) = (x)I is s aled identity metri ) 2014 Frank NIELSEN 5793b870 23/57
  • 24. Hyperboli Voronoi diagrams [19, 22℄ Hyperboli Voronoi diagram in Klein disk : 2014 Frank NIELSEN 5793b870 24/57
  • 25. Hyperboli Voronoi diagrams [19, 22℄ Ane Voronoi diagram equivalent to power diagram. Power distan e : kx − pk2 − wp (additively weighted ordinary Voronoi) 2014 Frank NIELSEN 5793b870 25/57
  • 26. Hyperboli Voronoi diagrams [19, 22℄ 5 ommon models of the abstra t hyperboli geometry https://www.youtube. om/wat h?v=i9IUzNxeH4o ACM Symposium on Computational Geometry (SoCG'14) 2014 Frank NIELSEN 5793b870 26/57
  • 27. k-order hyperboli Voronoi diagram k = 1 k = 2 Ane diagram ⇒ equivalent to a power diagram [21℄ Wat h k-order Voronoi : https://www.youtube. om/wat h?v=i9IUzNxeH4o 2014 Frank NIELSEN 5793b870 27/57
  • 28. Constrast fun tion and dually at spa e ◮ Convex and stri tly dierentiable fun tion F() admits a Legendre-Fen hel onvex onjugate F∗() : F∗() = sup (⊤ − F()), ∇F() = = (∇F∗)−1() ◮ Young's inequality gives rise to anoni al divergen e [8℄ : F()+F∗(′) ≥ ⊤′ ⇒ AF,F∗(, ′) = F() + F∗(′) − ⊤′ ◮ Writing using single oordinate system, get dual Bregman divergen es : BF (p : q) = F(p) − F(q) − (p − q)⊤∇F(q) = BF∗(q : p) = AF,F∗(p, q) = AF∗,F (q : p) 2014 Frank NIELSEN 5793b870 28/57
  • 29. Dis riminant analysis exponential families ◮ Consider X1 ∼ EF (1), X2 ∼ EF (2) (or moment parameterization i = ∇F(i )) ◮ Bayes' rule : Classify x from X1 i. P(x|1) P(x|2) ◮ Use bije tion between exponential families and Bregman divergen es : log p(x|) = −BF∗(t(x) : ) + F∗(t(x)) + k(x) ◮ MAP rule (w1 = w2 = 12 ) : BF∗(t(x) : 1) BF∗(t(x) : 2) ◮ Bregman bise tor : BiF∗(1, 2) = { |BF∗ ( : 1) = BF∗( : 1)} 2014 Frank NIELSEN 5793b870 29/57
  • 30. Bregman dual bise tors [3, 17, 20℄ Right-sided bise tor : → Hyperplane (-hyperplane) HF (p, q) = {x ∈ X | BF (x : p) = BF (x : q)}. HF : h∇F(p) − ∇F(q), xi + (F(p) − F(q) + hq,∇F(q)i − hp,∇F(p)i) = 0 Left-sided bise tor : → -Hypersurfa e (-hyperplane) H′F(p, q) = {x ∈ X | BF (p : x) = BF (q : x)} H′F : h∇F(x), q − pi + F(p) − F(q) = 0 2014 Frank NIELSEN 5793b870 30/57
  • 31. Visualizing Bregman bise tors Primal oordinates Dual oordinates natural parameters expe tation parameters p q Source Space: Itakura-Saito p(0.52977081,0.72041688) q(0.85824458,0.29083834) D(p,q)=0.66969016 D(q,p)=0.44835617 Gradient Space: Itakura-Saito dual p’(-1.88760873,-1.38808518) q’(-1.16516903,-3.43833618) D*(p’,q’)=0.44835617 D*(q’,p’)=0.66969016 p’ q’ 2014 Frank NIELSEN 5793b870 31/57
  • 32. Spa e of Bregman spheres and Bregman balls [3℄ Dual Bregman balls (bounding Bregman spheres) : Ballr F (c, r ) = {x ∈ X | BF (x : c) ≤ r} and Balll F (c, r ) = {x ∈ X | BF (c : x) ≤ r} Legendre duality : F (c, r ) = (∇F)−1(Ballr F∗(∇F(c), r )) Balll Illustration for Itakura-Saito divergen e, F(x) = −log x 2014 Frank NIELSEN 5793b870 32/57
  • 33. Lifting/Polarity : Potential fun tion graph F 2014 Frank NIELSEN 5793b870 33/57
  • 34. Spa e of Bregman spheres : Lifting map [3℄ F : x7→ ˆx = (x, F(x)), hypersurfa e in Rd+1. Hp : Tangent hyperplane at ˆp, z = Hp(x) = hx −p,∇F(p)i+F(p) ◮ Bregman sphere −→ ˆ with supporting hyperplane H : z = hx − c,∇F(c)i + F(c) + r . (// to Hc and shifted verti ally by r ) ˆ = F ∩ H. ◮ interse tion of any hyperplane H with F proje ts onto X as a Bregman sphere : H : z = hx, ai+b → : BallF (c = (∇F)−1(a), r = ha, ci−F(c)+b) 2014 Frank NIELSEN 5793b870 34/57
  • 35. Lifting with Itakura-Saito potential fun tion 2014 Frank NIELSEN 5793b870 35/57
  • 36. Spa e of Bregman spheres [3℄ ◮ Union/interse tion of Bregman spheres from representational polytope [3℄ ◮ Radi al axis of two Bregman balls is an hyperplane : Appli ations to Nearest Neighbor sear h trees like Bregman ball trees or Bregman vantage point trees [23℄. 2014 Frank NIELSEN 5793b870 36/57
  • 37. Spa e of spheres : Minimum en losing ball [14, 24℄ To a hyperplane H = H(a, b) : z = ha, xi+b in Rd+1, orresponds a ball = Ball(c, r ) in Rd with enter c = ∇F∗(a) and radius : r = ha, ci−F(c)+b = ha,∇F∗(a)i−F(∇F∗(a))+b = F∗(a)+b sin e F(∇F∗(a)) = h∇F∗(a), ai − F∗(a) (Young equality) SEB : Find halfspa e H(a, b)− : z ≤ ha, xi + b that ontains all lifted points : min a,b r = F∗(a) + b, ∀i ∈ {1, ..., n}, ha, xi i + b − F(xi ) ≥ 0. ⇒ Convex Program (CP) with linear inequality onstraints ⇒ F() = F∗() = 12 x⊤x : CP → Quadrati Programming (QP) [4℄. 2014 Frank NIELSEN 5793b870 37/57
  • 38. InSphere predi ates wrt Bregman divergen es [3℄ Impli it representation of Bregman spheres/balls : ◮ Is x inside the Bregman ball dened by d + 1 support points ? InSphere(x; p0, ..., pd ) =
  • 39.
  • 40.
  • 41.
  • 42.
  • 43.
  • 44. 1 ... 1 1 p0 ... pd x F(p0) ... F(pd ) F(x)
  • 45.
  • 46.
  • 47.
  • 48.
  • 49.
  • 50. ◮ InSphere(x; p0, ..., pd ) is negative, null or positive depending on whether x lies inside, on, or outside . 2014 Frank NIELSEN 5793b870 38/57
  • 51. Bregman Voronoi diagrams as minimization diagrams [3℄ A sub lass of ane diagrams whi h have all non-empty ells . Minimization diagram of the n fun tions Di (x) = BF (x : pi ) = F(x) − F(pi ) − hx − pi ,∇F(pi )i. ≡ minimization of n linear fun tions : Hi (x) = (pi − x)T∇F(qi ) − F(pi ) ⇐⇒ Wat h https://www.youtube. om/wat h?v=L7v1wuN_9Wg 2014 Frank NIELSEN 5793b870 39/57
  • 52. Bregman Voronoi from Power diagrams [15, 20℄ Any ane diagram an be built from a power diagram. (power diagrams dened in full spa e Rd ) ◮ Power distan e of x to Ball(p, r ) : ||p − x||2 − r 2. ◮ Laguerre diagram : minimization diagram of Di (x) = ||pi − x||2 − r 2 i ◮ Power bise tor of Ball(pi , ri ) and Ball(pj , rj )= radi al hyperplane : 2hx, pj − pi i + ||pi ||2 − ||pj ||2 + r 2 j − r 2 i = 0. Universality : Ane Bregman Voronoi diagram ⇔ Power diagram 2014 Frank NIELSEN 5793b870 40/57
  • 53. Ane Bregman Voronoi diagrams as power diagrams Equivalen e : B(∇F(pi ), ri ) with r 2 i = h∇F(pi ),∇F(pi )i + 2(F(pi ) − hpi ,∇F (pi )i (imaginary radii shown in red, WLOG. ri ≥ 0 by shifting) Some ells may be empty in the Laguerre diagram (power diagram) but not in the Bregman diagram http://www. sl.sony. o.jp/person/nielsen/BVDapplet/ 2014 Frank NIELSEN 5793b870 41/57
  • 54. Bregman dual Delaunay triangulations Delaunay Exponential Hellinger-like ◮ empty Bregman sphere property, ◮ geodesi triangles. BVDs extends Eu lidean Voronoi diagrams with similar omplexity/algorithms. 2014 Frank NIELSEN 5793b870 42/57
  • 55. Riemannian tensor and orthogonality Dually at spa e. ◮ F() onvex on yields onvex onjugate F∗() on H ◮ Same Riemannian tensor an be expressed in both oordinate systems : g() = ∇2F(), g∗() = ∇2F(), g((X))g∗((X)) = I , ∀X Same innitesimal length element ds2 = (ds∗)2. ◮ Two urves 1(t) and 2(t) are orthogonal at Q = 1(t0) = 2(t0) i. h ˙ 1(t0), ˙ 2(t0)ig((Q)) = h ˙ 1(t0), ˙ 2(t0)i∗g ∗((Q)) = 0 ◮ Geodesi (P,Q) and dual geodesi ∗(P,Q) (and geodesi segments [P,Q] and ∗[P,Q]) 2014 Frank NIELSEN 5793b870 43/57
  • 56. Orthogonality 3-point property (generalized law of osines) : BF (p : r ) = BF (p : q) + BF (q : r ) − (p − q)T (∇F(r ) − ∇F(q))) P Q R [P,Q] [Q,R] (P,Q) orthogonal to ∗(Q, R) i. BF (p : r ) = BF (p : q) + BF (q : r ) Equivalent to hp − q, r − qi = 0 Extend Pythagoras' theorem (P,Q) ⊥ ∗(Q, R) 2014 Frank NIELSEN 5793b870 44/57
  • 57. Dually orthogonal Bregman Voronoi Triangulations Ordinary Voronoi diagram is perpendi ular to Delaunay triangulation. (Vor k-fa e ⊥ Del d − k-fa e) Dual line segment geodesi s : (P,Q) = { = p + (1 − )q | ∈ [0, 1]} ∗(P,Q) = { = p + (1 − )q | ∈ [0, 1]} Bise tors : Bi(p, q) : hx, q − pi + F(p) − F(q) = 0 Bi(p, q) : hx, q − pi + F∗(p) − F∗(q) = 0 Dual orthogonality : Bi(p, q) ⊥ ∗(P,Q) (P,Q) ⊥ Bi(p, q) 2014 Frank NIELSEN 5793b870 45/57
  • 58. Dually orthogonal Bregman Voronoi Triangulations Bi(p, q) ⊥ ∗(P,Q) (P,Q) ⊥ Bi(p, q) 2014 Frank NIELSEN 5793b870 46/57
  • 59. MAP rule and additive Bregman Voronoi diagrams Optimal MAP de ision rule : map(x) = argmaxi∈{1,...,n}wipi (x) = argmaxi∈{1,...,n} − B∗(t(x) : i ) + log wi , = argmini∈{1,...,n}B∗(t(x) : i ) − log wi ⇒dual Bregman Voronoi with additive weights (ane diagram). Verti al proje tion of the interse tions of the lifted half-spa es on the potential F∗ shifted by the weights −log wi . 2014 Frank NIELSEN 5793b870 47/57
  • 60. Relative entropy for exponential families [18℄ ◮ Kullba k-Leibler divergen e ( ross-entropy minus entropy) : KL(P : Q) = Z p(x) log p(x) q(x) dx ≥ 0 = Z p(x) log 1 q(x) dx | {z } H×(P:Q)+c − Z p(x) log 1 p(x) dx | {z } H(p)=H×(P:P)−c = F(Q) − F(P) − hQ − P,∇F(P)i = BF (Q : P) = BF∗(P : Q) Bregman divergen e BF dened for a stri tly onvex and dierentiable fun tion up to some ane terms. ◮ Proof KL(P : Q) = BF (Q : P) follows from X ∼ EF () =⇒ E[t(X)] = ∇F() = 2014 Frank NIELSEN 5793b870 48/57
  • 61. Upper bounds Pe using Cherno Information [11℄ ◮ Tri k : min(a, b) ≤ ab1−, ∀ ∈ (0, 1) (for a, b 0) E = Z 1 w1− min(P(C1|x), P(C2|x))p(x)dx ≤ w 2 Z 1 (x)p1− p 2 (x)dx ◮ Upper bound the minimum error E∗ 1 w1− E∗ ≤ w 2 c(p1 : p2), c(p1 : p2) = R p 1 (x)p1− 2 (x)dx : Cherno - oe ient. c∗(p1 : p2) = c∗ (p1 : p2) = min ∈(0,1) Z p 1 (x)p1− 2 (x)dx. ◮ Cherno information (or Cherno divergen e) : C∗(p1 : p2) = C∗(p1 : p2) = max ∈(0,1) −log Z p 1 (x)p1− 2 (x)dx extend methodoloy with quasi-arithmeti means [11℄ 2014 Frank NIELSEN 5793b870 49/57
  • 62. Cherno oe ient/information : exponential families C(p : q) = −log c(p, q) = J() F (p : q), c(p : q) = e−C(p:q) = e−J() F (p:q). Skewed Jensen divergen e (on natural parameters) : J() F (p : q) = (F(p) + (1 − )F(q)) − F(p + (1 − )q) 2014 Frank NIELSEN 5793b870 50/57
  • 63. Maximizing skew Jensen divergen e ∗ = arg max 01 J() F (p : q) J(∗) F (p : q) = BF (p : m∗ ) = BF (q : m∗ ) m = p + (1 − )q : -mixing of p and q. J () BF (p : m F (p : q) ) BF (q : m) p q F(p) F(q) m = p + (1 − )q F Maximum skew Jensen divergen e amounts to Bregman divergen es. 2014 Frank NIELSEN 5793b870 51/57
  • 64. Geometry of the best error exponent : binary hypothesis [11℄ Cherno distribution P∗ : P∗ = P∗ 12 = Ge(P1, P2) ∩ Bim(P1, P2) e-geodesi : Ge(P1, P2) = {E() 12 | (E() 12 ) = (1 − )1 + 2, ∈ [0, 1]}, m-bise tor : Bim(P1, P2) : {P | F(1) − F(2) + (P)⊤ = 0}, Optimal natural parameter of P∗ : ∗ = (∗) 12 = argmin∈B(1 : ) = argmin∈B(2 : ). → losed-form for order-1 family, or e ient bise tion sear h. 2014 Frank NIELSEN 5793b870 52/57
  • 65. Geometry of the best error exponent : binary hypothesis P∗ = P∗ 12 = Ge(P1, P2) ∩ Bim(P1, P2) p1 p2 m-bisector p 12 -coordinate system e-geodesic Ge(P1 , P2 ) P 12 C(1 : 2) = B(1 : 12) Bim(P1 , P2 ) BHT : Pe bounded using Bregman divergen e between Cherno distribution and lass- onditional distributions. 2014 Frank NIELSEN 5793b870 53/57
  • 66. Geometry of the best error exponent : multiple hypothesis n-ary MHT [7℄ from minimum pairwise Cherno distan e : C(P1, ..., Pn) = min i ,j6=i C(Pi , Pj ) Pm e ≤ e−mC(Pi∗ ,Pj∗ ), (i ∗, j∗) = argmini ,j6=iC(Pi , Pj ) Compute for ea h pair of natural neighbors Pi and Pj , the Cherno distan e C(Pi , Pj ), and hoose the pair with minimal distan e. (Proof by ontradi tion using Bregman Pythagoras theorem.) 2014 Frank NIELSEN 5793b870 54/57
  • 67. Multiple Hypothesis testing Cherno information -coordinate system Chernoff distribution between natural neighbours 2014 Frank NIELSEN 5793b870 55/57
  • 68. Summary : Statisti al Voronoi diagrams ◮ Mahalanobis Voronoi diagrams, anisotropi Voronoi diagrams (additively-weighted) ◮ Fisher-Hotelling-Rao Riemannian Voronoi diagrams ◮ Kullba k-Leibler Voronoi diagrams for exponential families = Bregman Voronoi diagrams ◮ Extend ordinary Voronoi diagrams (≡ isotropi Gaussians) ◮ Spa e of spheres : Lifting/polarity ◮ bise tors ⊥ geodesi s ◮ dual regular triangulation with empty spheres Information geometry : ane dierential geometry of parameter spa es, invarian e prin iples Many other kinds of Voronoi : Jensen-Bregman, (u, v)-stru tures , onformal divergen es, total Bregman total Jensen divergen es, et . 2014 Frank NIELSEN 5793b870 56/57
  • 69. Computational Information Geometry : Edited books [13℄ [12℄ http://www.springer. om/engineering/signals/book/978-3-642-30231-2 http://www.sony sl. o.jp/person/nielsen/infogeo/MIG/MIGBOOKWEB/ http://www.springer. om/engineering/signals/book/978-3-319-05316-5 http://www.sony sl. o.jp/person/nielsen/infogeo/GTI/Geometri TheoryOfInformation.html 2014 Frank NIELSEN 5793b870 57/57
  • 70. Mar Arnaudon and Frank Nielsen. On approximating the Riemannian 1- enter. Comput. Geom. Theory Appl., 46(1) :93104, January 2013. Jean-Daniel Boissonnat and Christophe Delage. Convex hull and Voronoi diagram of additively weighted points. In Gerth St ¸lting Brodal and Stefano Leonardi, editors, ESA, volume 3669 of Le ture Notes in Computer S ien e, pages 367378. Springer, 2005. Jean-Daniel Boissonnat, Frank Nielsen, and Ri hard No k. Bregman Voronoi diagrams. Dis rete and Computational Geometry, 44(2) :281307, April 2010. Bernd Gärtner and Sven S hönherr. An e ient, exa t, and generi quadrati programming solver for geometri optimization. In Pro eedings of the sixteenth annual symposium on Computational geometry, pages 110118. ACM, 2000. Harold Hotelling. Fran ois Labelle and Jonathan Ri hard Shew huk. Anisotropi Voronoi diagrams and guaranteed-quality anisotropi mesh generation. In Pro eedings of the nineteenth annual symposium on Computational geometry, pages 191200. ACM, 2003. C. C. Leang and D. H. Johnson. On the asymptoti s of M-hypothesis Bayesian dete tion. IEEE Transa tions on Information Theory, 43(1) :280282, January 1997. Frank Nielsen. Legendre transformation and information geometry. Te hni al Report CIG-MEMO2, September 2010. Frank Nielsen. Hypothesis testing, information divergen e and omputational geometry. 2014 Frank NIELSEN 5793b870 57/57
  • 71. In Frank Nielsen and Fr ©d ©ri Barbares o, editors, Geometri S ien e of Information, volume 8085 of Le ture Notes in Computer S ien e, pages 241248. Springer Berlin Heidelberg, 2013. Frank Nielsen. Generalized bhatta haryya and herno upper bounds on bayes error using quasi-arithmeti means. Pattern Re ognition Letters, 42 :2534, 2014. Frank Nielsen. Generalized bhatta haryya and herno upper bounds on bayes error using quasi-arithmeti means. Pattern Re ognition Letters, 42 :2534, 2014. Frank Nielsen. Geometri Theory of Information. Springer, 2014. Frank Nielsen and Rajendra Bhatia, editors. Matrix Information Geometry (Revised Invited Papers). Springer, 2012. Frank Nielsen and Ri hard No k. On the smallest en losing information disk. Information Pro essing Letters (IPL), 105(3) :9397, 2008. Frank Nielsen and Ri hard No k. Quantum Voronoi diagrams and Holevo hannel apa ity for 1-qubit quantum states. In IEEE International Symposium on Information Theory (ISIT), pages 96100, Toronto, Canada, July 2008. IEEE. Frank Nielsen and Ri hard No k. Approximating smallest en losing balls with appli ations to ma hine learning. Int. J. Comput. Geometry Appl., 19(5) :389414, 2009. Frank Nielsen and Ri hard No k. The dual Voronoi diagrams with respe t to representational Bregman divergen es. In International Symposium on Voronoi Diagrams (ISVD), pages 7178, 2009. 2014 Frank NIELSEN 5793b870 57/57
  • 72. Frank Nielsen and Ri hard No k. Entropies and ross-entropies of exponential families. In Pro eedings of the International Conferen e on Image Pro essing, ICIP 2010, September 26-29, Hong Kong, China, pages 36213624, 2010. Frank Nielsen and Ri hard No k. Hyperboli Voronoi diagrams made easy. In 2013 13th International Conferen e on Computational S ien e and Its Appli ations, pages 7480. IEEE, 2010. Frank Nielsen and Ri hard No k. Hyperboli Voronoi diagrams made easy. In International Conferen e on Computational S ien e and its Appli ations (ICCSA), volume 1, pages 7480, Los Alamitos, CA, USA, mar h 2010. IEEE Computer So iety. Frank Nielsen and Ri hard No k. Further results on the hyperboli Voronoi diagrams. ISVD, 2014. Frank Nielsen and Ri hard No k. Visualizing hyperboli Voronoi diagrams. In Symposium on Computational Geometry, page 90, 2014. Frank Nielsen, Paolo Piro, and Mi hel Barlaud. Bregman vantage point trees for e ient nearest neighbor queries. In Pro eedings of the 2009 IEEE International Conferen e on Multimedia and Expo (ICME), pages 878881, 2009. Ri hard No k and Frank Nielsen. Fitting the smallest en losing bregman balls. In 16th European Conferen e on Ma hine Learning (ECML), pages 649656, O tober 2005. Calyampudi Radhakrishna Rao. Information and the a ura y attainable in the estimation of statisti al parameters. Bulletin of the Cal utta Mathemati al So iety, 37 :8189, 1945. 2014 Frank NIELSEN 5793b870 57/57