rtugal o P · rt pa of assessing the y qualit a given rtition. pa Dunn's index and e Dunn-lik...
Transcript of rtugal o P · rt pa of assessing the y qualit a given rtition. pa Dunn's index and e Dunn-lik...
A general notion of interiority 6th Winter S hool on Te hnology Assessment � 1
A general notion of interiority
J. Orestes Cerdeira
Fa uldade de Ciên ias e Te nologia,
Universidade Nova de Lisboa,
Portugal
Joint with M. João Martins, Pedro C. Silva and T. Monteiro-Henriques
two issues...
⊲ Introdu tion
Interiority
Separability index
Convexity riterion
Auto-interiority
A general notion of interiority 6th Winter S hool on Te hnology Assessment � 2
to what extent a set of entities is separated from ?
how are the entities of a group arranged together ?
(more on entrated to the "interior� or dispersed on the
"margins� of their range)
two issues...
⊲ Introdu tion
Interiority
Separability index
Convexity riterion
Auto-interiority
A general notion of interiority 6th Winter S hool on Te hnology Assessment � 2
� to what extent a set of entities {•} is separated from • ?
how are the entities of a group arranged together ?
(more on entrated to the "interior� or dispersed on the
"margins� of their range)
two issues...
⊲ Introdu tion
Interiority
Separability index
Convexity riterion
Auto-interiority
A general notion of interiority 6th Winter S hool on Te hnology Assessment � 2
� to what extent a set of entities {•} is separated from • ?
how are the entities of a group arranged together ?
(more on entrated to the "interior� or dispersed on the
"margins� of their range)
two issues...
⊲ Introdu tion
Interiority
Separability index
Convexity riterion
Auto-interiority
A general notion of interiority 6th Winter S hool on Te hnology Assessment � 2
� to what extent a set of entities {•} is separated from • ?
how are the entities of a group arranged together ?
(more on entrated to the "interior� or dispersed on the
"margins� of their range)
two issues...
⊲ Introdu tion
Interiority
Separability index
Convexity riterion
Auto-interiority
A general notion of interiority 6th Winter S hool on Te hnology Assessment � 2
� to what extent a set of entities {•} is separated from • ?
� how are the entities of a group arranged together ?
(more on entrated to the "interior� or dispersed on the
"margins� of their range)
two issues...
⊲ Introdu tion
Interiority
Separability index
Convexity riterion
Auto-interiority
A general notion of interiority 6th Winter S hool on Te hnology Assessment � 2
� to what extent a set of entities {•} is separated from • ?
� how are the entities of a group arranged together ?
(more on entrated to the "interior� or dispersed on the
"margins� of their range)
two issues...
⊲ Introdu tion
Interiority
Separability index
Convexity riterion
Auto-interiority
A general notion of interiority 6th Winter S hool on Te hnology Assessment � 2
� to what extent a set of entities {•} is separated from • ?
� how are the entities of a group arranged together ?
(more on entrated to the "interior� or dispersed on the
"margins� of their range)
two issues...
⊲ Introdu tion
Interiority
Separability index
Convexity riterion
Auto-interiority
A general notion of interiority 6th Winter S hool on Te hnology Assessment � 2
� to what extent a set of entities {•} is separated from • ?
� how are the entities of a group arranged together ?
(more on entrated to the "interior� or dispersed on the
"margins� of their range)
issue 1: is {•} well separated from {•}?
⊲ Introdu tion
Interiority
Separability index
Convexity riterion
Auto-interiority
A general notion of interiority 6th Winter S hool on Te hnology Assessment � 3
issue 1: is {•} well separated from {•}?
⊲ Introdu tion
Interiority
Separability index
Convexity riterion
Auto-interiority
A general notion of interiority 6th Winter S hool on Te hnology Assessment � 3
In many areas ...
⊲ Introdu tion
Interiority
Separability index
Convexity riterion
Auto-interiority
A general notion of interiority 6th Winter S hool on Te hnology Assessment � 4
E ology, Chemistry, Mole ular Biology, Physiology, Toxi ology,
So ial S ien es ...
issue: to what extent a parti ular subset of entities is
separated from the other entities? arises
In many areas ...
⊲ Introdu tion
Interiority
Separability index
Convexity riterion
Auto-interiority
A general notion of interiority 6th Winter S hool on Te hnology Assessment � 4
E ology, Chemistry, Mole ular Biology, Physiology, Toxi ology,
So ial S ien es ...
issue: to what extent a parti ular subset of entities is
separated from the other entities? arises
In many areas ...
⊲ Introdu tion
Interiority
Separability index
Convexity riterion
Auto-interiority
A general notion of interiority 6th Winter S hool on Te hnology Assessment � 4
E ology, Chemistry, Mole ular Biology, Physiology, Toxi ology,
So ial S ien es ...
issue:
to what extent a parti ular subset of entities is
separated from the other entities? arises
In many areas ...
⊲ Introdu tion
Interiority
Separability index
Convexity riterion
Auto-interiority
A general notion of interiority 6th Winter S hool on Te hnology Assessment � 4
E ology, Chemistry, Mole ular Biology, Physiology, Toxi ology,
So ial S ien es ...
issue: to what extent a parti ular subset X of entities is
separated from the other entities?
arises
In many areas ...
⊲ Introdu tion
Interiority
Separability index
Convexity riterion
Auto-interiority
A general notion of interiority 6th Winter S hool on Te hnology Assessment � 4
E ology, Chemistry, Mole ular Biology, Physiology, Toxi ology,
So ial S ien es ...
issue: to what extent a parti ular subset X of entities is
separated from the other entities? arises
In the ontext of luster analysis...
⊲ Introdu tion
Interiority
Separability index
Convexity riterion
Auto-interiority
A general notion of interiority 6th Winter S hool on Te hnology Assessment � 5
methods have been proposed that measure overlap/separability of
lusters as part of assessing the quality of a given partition.
Dunn's index and Dunn-like indi es, Davies-Bouldin's index, the
silhouette index and Thornton's index.
With these approa hes separability is numeri ally quanti�ed, but
no detailed information on how ea h luster is separated from
the others is produ ed
all, ex ept Thornton's index, are expressions that expli itly
involve numeri al dissimilarities between entities
notions su h as �perfe t separation of luster� are not
omprise.
In the ontext of luster analysis...
⊲ Introdu tion
Interiority
Separability index
Convexity riterion
Auto-interiority
A general notion of interiority 6th Winter S hool on Te hnology Assessment � 5
methods have been proposed that measure overlap/separability of
lusters as part of assessing the quality of a given partition.
Dunn's index and Dunn-like indi es, Davies-Bouldin's index, the
silhouette index and Thornton's index.
With these approa hes separability is numeri ally quanti�ed, but
no detailed information on how ea h luster is separated from
the others is produ ed
all, ex ept Thornton's index, are expressions that expli itly
involve numeri al dissimilarities between entities
notions su h as �perfe t separation of luster� are not
omprise.
In the ontext of luster analysis...
⊲ Introdu tion
Interiority
Separability index
Convexity riterion
Auto-interiority
A general notion of interiority 6th Winter S hool on Te hnology Assessment � 5
methods have been proposed that measure overlap/separability of
lusters as part of assessing the quality of a given partition.
Dunn's index and Dunn-like indi es, Davies-Bouldin's index, the
silhouette index and Thornton's index.
With these approa hes separability is numeri ally quanti�ed
, but
no detailed information on how ea h luster is separated from
the others is produ ed
all, ex ept Thornton's index, are expressions that expli itly
involve numeri al dissimilarities between entities
notions su h as �perfe t separation of luster� are not
omprise.
In the ontext of luster analysis...
⊲ Introdu tion
Interiority
Separability index
Convexity riterion
Auto-interiority
A general notion of interiority 6th Winter S hool on Te hnology Assessment � 5
methods have been proposed that measure overlap/separability of
lusters as part of assessing the quality of a given partition.
Dunn's index and Dunn-like indi es, Davies-Bouldin's index, the
silhouette index and Thornton's index.
With these approa hes separability is numeri ally quanti�ed, but
no detailed information on how ea h luster is separated from
the others is produ ed
all, ex ept Thornton's index, are expressions that expli itly
involve numeri al dissimilarities between entities
notions su h as �perfe t separation of luster� are not
omprise.
In the ontext of luster analysis...
⊲ Introdu tion
Interiority
Separability index
Convexity riterion
Auto-interiority
A general notion of interiority 6th Winter S hool on Te hnology Assessment � 5
methods have been proposed that measure overlap/separability of
lusters as part of assessing the quality of a given partition.
Dunn's index and Dunn-like indi es, Davies-Bouldin's index, the
silhouette index and Thornton's index.
With these approa hes separability is numeri ally quanti�ed, but
� no detailed information on how ea h luster is separated from
the others is produ ed
all, ex ept Thornton's index, are expressions that expli itly
involve numeri al dissimilarities between entities
notions su h as �perfe t separation of luster� are not
omprise.
In the ontext of luster analysis...
⊲ Introdu tion
Interiority
Separability index
Convexity riterion
Auto-interiority
A general notion of interiority 6th Winter S hool on Te hnology Assessment � 5
methods have been proposed that measure overlap/separability of
lusters as part of assessing the quality of a given partition.
Dunn's index and Dunn-like indi es, Davies-Bouldin's index, the
silhouette index and Thornton's index.
With these approa hes separability is numeri ally quanti�ed, but
� no detailed information on how ea h luster is separated from
the others is produ ed
� all, ex ept Thornton's index, are expressions that expli itly
involve numeri al dissimilarities between entities
notions su h as �perfe t separation of luster� are not
omprise.
In the ontext of luster analysis...
⊲ Introdu tion
Interiority
Separability index
Convexity riterion
Auto-interiority
A general notion of interiority 6th Winter S hool on Te hnology Assessment � 5
methods have been proposed that measure overlap/separability of
lusters as part of assessing the quality of a given partition.
Dunn's index and Dunn-like indi es, Davies-Bouldin's index, the
silhouette index and Thornton's index.
With these approa hes separability is numeri ally quanti�ed, but
� no detailed information on how ea h luster is separated from
the others is produ ed
� all, ex ept Thornton's index, are expressions that expli itly
involve numeri al dissimilarities between entities
� notions su h as �perfe t separation of luster� are not
omprise.
Silhouette index
⊲ Introdu tion
Interiority
Separability index
Convexity riterion
Auto-interiority
A general notion of interiority 6th Winter S hool on Te hnology Assessment � 6
To onstru t the silhouette S(X) (with respe t to X̄), for i ∈ X
S(i) =b(i)− a(i)
max{a(i), b(i)}
a(i) - average dissimilarity of entity i to all other entities in X
b(i) - average dissimilarity of entity i to all entities in X̄
.
, means is �well- lustered� in
, means lies �equally far away� from and
, means is �mis lassi�ed� as an entity.
is the average of the silhouettes of
notions su h as �perfe t separation of luster� are not omprise.
Silhouette index
⊲ Introdu tion
Interiority
Separability index
Convexity riterion
Auto-interiority
A general notion of interiority 6th Winter S hool on Te hnology Assessment � 6
To onstru t the silhouette S(X) (with respe t to X̄), for i ∈ X
S(i) =b(i)− a(i)
max{a(i), b(i)}
a(i) - average dissimilarity of entity i to all other entities in X
b(i) - average dissimilarity of entity i to all entities in X̄
S(i) ∈ [−1, 1].
, means is �well- lustered� in
, means lies �equally far away� from and
, means is �mis lassi�ed� as an entity.
is the average of the silhouettes of
notions su h as �perfe t separation of luster� are not omprise.
Silhouette index
⊲ Introdu tion
Interiority
Separability index
Convexity riterion
Auto-interiority
A general notion of interiority 6th Winter S hool on Te hnology Assessment � 6
To onstru t the silhouette S(X) (with respe t to X̄), for i ∈ X
S(i) =b(i)− a(i)
max{a(i), b(i)}
a(i) - average dissimilarity of entity i to all other entities in X
b(i) - average dissimilarity of entity i to all entities in X̄
S(i) ∈ [−1, 1].
S(i) ≃ 1, means i is �well- lustered� in X
, means lies �equally far away� from and
, means is �mis lassi�ed� as an entity.
is the average of the silhouettes of
notions su h as �perfe t separation of luster� are not omprise.
Silhouette index
⊲ Introdu tion
Interiority
Separability index
Convexity riterion
Auto-interiority
A general notion of interiority 6th Winter S hool on Te hnology Assessment � 6
To onstru t the silhouette S(X) (with respe t to X̄), for i ∈ X
S(i) =b(i)− a(i)
max{a(i), b(i)}
a(i) - average dissimilarity of entity i to all other entities in X
b(i) - average dissimilarity of entity i to all entities in X̄
S(i) ∈ [−1, 1].
S(i) ≃ 1, means i is �well- lustered� in X
S(i) ≃ 0, means i lies �equally far away� from X and X̄
, means is �mis lassi�ed� as an entity.
is the average of the silhouettes of
notions su h as �perfe t separation of luster� are not omprise.
Silhouette index
⊲ Introdu tion
Interiority
Separability index
Convexity riterion
Auto-interiority
A general notion of interiority 6th Winter S hool on Te hnology Assessment � 6
To onstru t the silhouette S(X) (with respe t to X̄), for i ∈ X
S(i) =b(i)− a(i)
max{a(i), b(i)}
a(i) - average dissimilarity of entity i to all other entities in X
b(i) - average dissimilarity of entity i to all entities in X̄
S(i) ∈ [−1, 1].
S(i) ≃ 1, means i is �well- lustered� in X
S(i) ≃ 0, means i lies �equally far away� from X and X̄
S(i) ≃ −1, means i is �mis lassi�ed� as an X entity.
is the average of the silhouettes of
notions su h as �perfe t separation of luster� are not omprise.
Silhouette index
⊲ Introdu tion
Interiority
Separability index
Convexity riterion
Auto-interiority
A general notion of interiority 6th Winter S hool on Te hnology Assessment � 6
To onstru t the silhouette S(X) (with respe t to X̄), for i ∈ X
S(i) =b(i)− a(i)
max{a(i), b(i)}
a(i) - average dissimilarity of entity i to all other entities in X
b(i) - average dissimilarity of entity i to all entities in X̄
S(i) ∈ [−1, 1].
S(i) ≃ 1, means i is �well- lustered� in X
S(i) ≃ 0, means i lies �equally far away� from X and X̄
S(i) ≃ −1, means i is �mis lassi�ed� as an X entity.
S(X) is the average of the silhouettes of i ∈ X
notions su h as �perfe t separation of luster� are not omprise.
Silhouette index
⊲ Introdu tion
Interiority
Separability index
Convexity riterion
Auto-interiority
A general notion of interiority 6th Winter S hool on Te hnology Assessment � 6
To onstru t the silhouette S(X) (with respe t to X̄), for i ∈ X
S(i) =b(i)− a(i)
max{a(i), b(i)}
a(i) - average dissimilarity of entity i to all other entities in X
b(i) - average dissimilarity of entity i to all entities in X̄
S(i) ∈ [−1, 1].
S(i) ≃ 1, means i is �well- lustered� in X
S(i) ≃ 0, means i lies �equally far away� from X and X̄
S(i) ≃ −1, means i is �mis lassi�ed� as an X entity.
S(X) is the average of the silhouettes of i ∈ X
notions su h as �perfe t separation of luster� are not omprise.
Thornton's index
⊲ Introdu tion
Interiority
Separability index
Convexity riterion
Auto-interiority
A general notion of interiority 6th Winter S hool on Te hnology Assessment � 7
if the nearest neighbor of is (not) in
Thornton's index
⊲ Introdu tion
Interiority
Separability index
Convexity riterion
Auto-interiority
A general notion of interiority 6th Winter S hool on Te hnology Assessment � 7
GSI =
∑i∈X δi
|X|, δi = 1(0) if the nearest neighbor of i is (not) in X
Thornton's index
⊲ Introdu tion
Interiority
Separability index
Convexity riterion
Auto-interiority
A general notion of interiority 6th Winter S hool on Te hnology Assessment � 7
GSI =
∑i∈X δi
|X|, δi = 1(0) if the nearest neighbor of i is (not) in X
Thornton's index
⊲ Introdu tion
Interiority
Separability index
Convexity riterion
Auto-interiority
A general notion of interiority 6th Winter S hool on Te hnology Assessment � 7
GSI =
∑i∈X δi
|X|, δi = 1(0) if the nearest neighbor of i is (not) in X
GSI = 1
Thornton's index
⊲ Introdu tion
Interiority
Separability index
Convexity riterion
Auto-interiority
A general notion of interiority 6th Winter S hool on Te hnology Assessment � 7
GSI =
∑i∈X δi
|X|, δi = 1(0) if the nearest neighbor of i is (not) in X
GSI = 1
Thornton's index
⊲ Introdu tion
Interiority
Separability index
Convexity riterion
Auto-interiority
A general notion of interiority 6th Winter S hool on Te hnology Assessment � 7
GSI =
∑i∈X δi
|X|, δi = 1(0) if the nearest neighbor of i is (not) in X
GSI = 1
GSI = 1
other proposals rely on a notion of territory
⊲ Introdu tion
Interiority
Separability index
Convexity riterion
Auto-interiority
A general notion of interiority 6th Winter S hool on Te hnology Assessment � 8
doesn't dis riminate separability
other proposals rely on a notion of territory
⊲ Introdu tion
Interiority
Separability index
Convexity riterion
Auto-interiority
A general notion of interiority 6th Winter S hool on Te hnology Assessment � 8
doesn't dis riminate separability
other proposals rely on a notion of territory
⊲ Introdu tion
Interiority
Separability index
Convexity riterion
Auto-interiority
A general notion of interiority 6th Winter S hool on Te hnology Assessment � 8
doesn't dis riminate separability
other proposals rely on a notion of territory
⊲ Introdu tion
Interiority
Separability index
Convexity riterion
Auto-interiority
A general notion of interiority 6th Winter S hool on Te hnology Assessment � 8
doesn't dis riminate separability
Instead of deeming...
⊲ Introdu tion
Interiority
Separability index
Convexity riterion
Auto-interiority
A general notion of interiority 6th Winter S hool on Te hnology Assessment � 9
to what extent a set of entities X is separated from other
entities?
to what extent a set of entities is separated from another
entity?
Instead of deeming...
⊲ Introdu tion
Interiority
Separability index
Convexity riterion
Auto-interiority
A general notion of interiority 6th Winter S hool on Te hnology Assessment � 9
to what extent a set of entities X is separated from other
entities?
to what extent a set of entities X is separated from another
entity?
to what extend {•} is separated from • ?
⊲ Introdu tion
Interiority
Separability index
Convexity riterion
Auto-interiority
A general notion of interiority 6th Winter S hool on Te hnology Assessment � 10
to what extend {•} is separated from • ?
⊲ Introdu tion
Interiority
Separability index
Convexity riterion
Auto-interiority
A general notion of interiority 6th Winter S hool on Te hnology Assessment � 10
to what extend {•} is separated from • ?
⊲ Introdu tion
Interiority
Separability index
Convexity riterion
Auto-interiority
A general notion of interiority 6th Winter S hool on Te hnology Assessment � 10
to what extend {•} is separated from • ?
⊲ Introdu tion
Interiority
Separability index
Convexity riterion
Auto-interiority
A general notion of interiority 6th Winter S hool on Te hnology Assessment � 10
to what extend {•} is separated from • ?
⊲ Introdu tion
Interiority
Separability index
Convexity riterion
Auto-interiority
A general notion of interiority 6th Winter S hool on Te hnology Assessment � 10
to formalize interiority needs �xing some aggregation riterion
⊲ Introdu tion
Interiority
Separability index
Convexity riterion
Auto-interiority
A general notion of interiority 6th Winter S hool on Te hnology Assessment � 11
i.e., a rule that identi�es, for a set of entities and ,
the -partitions of whi h are adequately aggregated
is adequate
if the sum of distan es from ea h point to its luster enter is
minimum ( -means)
if the largest diameter of its lusters is minimum (diameter)
if lusters are the onne ted omponents of some minimum
-spanning forest (single linkage)
to formalize interiority needs �xing some aggregation riterion
⊲ Introdu tion
Interiority
Separability index
Convexity riterion
Auto-interiority
A general notion of interiority 6th Winter S hool on Te hnology Assessment � 11
i.e., a rule that identi�es, for a set of entities E and 1 ≤ k ≤ |E|,
the k-partitions of E whi h are adequately aggregated
is adequate
if the sum of distan es from ea h point to its luster enter is
minimum ( -means)
if the largest diameter of its lusters is minimum (diameter)
if lusters are the onne ted omponents of some minimum
-spanning forest (single linkage)
to formalize interiority needs �xing some aggregation riterion
⊲ Introdu tion
Interiority
Separability index
Convexity riterion
Auto-interiority
A general notion of interiority 6th Winter S hool on Te hnology Assessment � 11
i.e., a rule that identi�es, for a set of entities E and 1 ≤ k ≤ |E|,
the k-partitions of E whi h are adequately aggregated
Pk is adequate
if the sum of distan es from ea h point to its luster enter is
minimum ( -means)
if the largest diameter of its lusters is minimum (diameter)
if lusters are the onne ted omponents of some minimum
-spanning forest (single linkage)
to formalize interiority needs �xing some aggregation riterion
⊲ Introdu tion
Interiority
Separability index
Convexity riterion
Auto-interiority
A general notion of interiority 6th Winter S hool on Te hnology Assessment � 11
i.e., a rule that identi�es, for a set of entities E and 1 ≤ k ≤ |E|,
the k-partitions of E whi h are adequately aggregated
Pk is adequate
� if the sum of distan es from ea h point to its luster enter is
minimum
( -means)
if the largest diameter of its lusters is minimum (diameter)
if lusters are the onne ted omponents of some minimum
-spanning forest (single linkage)
to formalize interiority needs �xing some aggregation riterion
⊲ Introdu tion
Interiority
Separability index
Convexity riterion
Auto-interiority
A general notion of interiority 6th Winter S hool on Te hnology Assessment � 11
i.e., a rule that identi�es, for a set of entities E and 1 ≤ k ≤ |E|,
the k-partitions of E whi h are adequately aggregated
Pk is adequate
� if the sum of distan es from ea h point to its luster enter is
minimum (k-means)
if the largest diameter of its lusters is minimum (diameter)
if lusters are the onne ted omponents of some minimum
-spanning forest (single linkage)
to formalize interiority needs �xing some aggregation riterion
⊲ Introdu tion
Interiority
Separability index
Convexity riterion
Auto-interiority
A general notion of interiority 6th Winter S hool on Te hnology Assessment � 11
i.e., a rule that identi�es, for a set of entities E and 1 ≤ k ≤ |E|,
the k-partitions of E whi h are adequately aggregated
Pk is adequate
� if the sum of distan es from ea h point to its luster enter is
minimum (k-means)
� if the largest diameter of its lusters is minimum
(diameter)
if lusters are the onne ted omponents of some minimum
-spanning forest (single linkage)
to formalize interiority needs �xing some aggregation riterion
⊲ Introdu tion
Interiority
Separability index
Convexity riterion
Auto-interiority
A general notion of interiority 6th Winter S hool on Te hnology Assessment � 11
i.e., a rule that identi�es, for a set of entities E and 1 ≤ k ≤ |E|,
the k-partitions of E whi h are adequately aggregated
Pk is adequate
� if the sum of distan es from ea h point to its luster enter is
minimum (k-means)
� if the largest diameter of its lusters is minimum (diameter)
if lusters are the onne ted omponents of some minimum
-spanning forest (single linkage)
to formalize interiority needs �xing some aggregation riterion
⊲ Introdu tion
Interiority
Separability index
Convexity riterion
Auto-interiority
A general notion of interiority 6th Winter S hool on Te hnology Assessment � 11
i.e., a rule that identi�es, for a set of entities E and 1 ≤ k ≤ |E|,
the k-partitions of E whi h are adequately aggregated
Pk is adequate
� if the sum of distan es from ea h point to its luster enter is
minimum (k-means)
� if the largest diameter of its lusters is minimum (diameter)
� if lusters are the onne ted omponents of some minimum
(|E| − k)-spanning forest
(single linkage)
to formalize interiority needs �xing some aggregation riterion
⊲ Introdu tion
Interiority
Separability index
Convexity riterion
Auto-interiority
A general notion of interiority 6th Winter S hool on Te hnology Assessment � 11
i.e., a rule that identi�es, for a set of entities E and 1 ≤ k ≤ |E|,
the k-partitions of E whi h are adequately aggregated
Pk is adequate
� if the sum of distan es from ea h point to its luster enter is
minimum (k-means)
� if the largest diameter of its lusters is minimum (diameter)
� if lusters are the onne ted omponents of some minimum
(|E| − k)-spanning forest (single linkage)
to formalize interiority needs �xing some aggregation riterion
⊲ Introdu tion
Interiority
Separability index
Convexity riterion
Auto-interiority
A general notion of interiority 6th Winter S hool on Te hnology Assessment � 11
i.e., a rule that identi�es, for a set of entities E and 1 ≤ k ≤ |E|,
the k-partitions of E whi h are adequately aggregated
Pk is adequate
� if the sum of distan es from ea h point to its luster enter is
minimum (k-means)
� if the largest diameter of its lusters is minimum (diameter)
� if lusters are the onne ted omponents of some minimum
(|E| − k)-spanning forest (single linkage)
� · · ·
k-interior
Introdu tion
⊲ Interiority
Separability index
Convexity riterion
Auto-interiority
A general notion of interiority 6th Winter S hool on Te hnology Assessment � 12
entity x̄ is k-interior with respe t to X if for every k-partition Pk
of X, Pk ∪{{x̄}} is not an adequate (k+1)-partition of X ∪{x̄}
is 1-interior
is 2-interior
k-interior
Introdu tion
⊲ Interiority
Separability index
Convexity riterion
Auto-interiority
A general notion of interiority 6th Winter S hool on Te hnology Assessment � 12
entity x̄ is k-interior with respe t to X if for every k-partition Pk
of X, Pk ∪{{x̄}} is not an adequate (k+1)-partition of X ∪{x̄}
is
1-interior
is 2-interior
k-interior
Introdu tion
⊲ Interiority
Separability index
Convexity riterion
Auto-interiority
A general notion of interiority 6th Winter S hool on Te hnology Assessment � 12
entity x̄ is k-interior with respe t to X if for every k-partition Pk
of X, Pk ∪{{x̄}} is not an adequate (k+1)-partition of X ∪{x̄}
is
1-interior
is 2-interior
k-interior
Introdu tion
⊲ Interiority
Separability index
Convexity riterion
Auto-interiority
A general notion of interiority 6th Winter S hool on Te hnology Assessment � 12
entity x̄ is k-interior with respe t to X if for every k-partition Pk
of X, Pk ∪{{x̄}} is not an adequate (k+1)-partition of X ∪{x̄}
is 1-interior
is 2-interior
k-interior
Introdu tion
⊲ Interiority
Separability index
Convexity riterion
Auto-interiority
A general notion of interiority 6th Winter S hool on Te hnology Assessment � 12
entity x̄ is k-interior with respe t to X if for every k-partition Pk
of X, Pk ∪{{x̄}} is not an adequate (k+1)-partition of X ∪{x̄}
is
1-interior
is 2-interior
k-interior
Introdu tion
⊲ Interiority
Separability index
Convexity riterion
Auto-interiority
A general notion of interiority 6th Winter S hool on Te hnology Assessment � 12
entity x̄ is k-interior with respe t to X if for every k-partition Pk
of X, Pk ∪{{x̄}} is not an adequate (k+1)-partition of X ∪{x̄}
is 1-interior
is 2-interior
k-interior
Introdu tion
⊲ Interiority
Separability index
Convexity riterion
Auto-interiority
A general notion of interiority 6th Winter S hool on Te hnology Assessment � 12
entity x̄ is k-interior with respe t to X if for every k-partition Pk
of X, Pk ∪{{x̄}} is not an adequate (k+1)-partition of X ∪{x̄}
• is 1-interior
is 2-interior
k-interior
Introdu tion
⊲ Interiority
Separability index
Convexity riterion
Auto-interiority
A general notion of interiority 6th Winter S hool on Te hnology Assessment � 12
entity x̄ is k-interior with respe t to X if for every k-partition Pk
of X, Pk ∪{{x̄}} is not an adequate (k+1)-partition of X ∪{x̄}
• is 1-interior
is 2-interior
k-interior
Introdu tion
⊲ Interiority
Separability index
Convexity riterion
Auto-interiority
A general notion of interiority 6th Winter S hool on Te hnology Assessment � 12
entity x̄ is k-interior with respe t to X if for every k-partition Pk
of X, Pk ∪{{x̄}} is not an adequate (k+1)-partition of X ∪{x̄}
• is 1-interior
is 2-interior
k-interior
Introdu tion
⊲ Interiority
Separability index
Convexity riterion
Auto-interiority
A general notion of interiority 6th Winter S hool on Te hnology Assessment � 12
entity x̄ is k-interior with respe t to X if for every k-partition Pk
of X, Pk ∪{{x̄}} is not an adequate (k+1)-partition of X ∪{x̄}
• is 1-interior
is 2-interior
k-interior
Introdu tion
⊲ Interiority
Separability index
Convexity riterion
Auto-interiority
A general notion of interiority 6th Winter S hool on Te hnology Assessment � 12
entity x̄ is k-interior with respe t to X if for every k-partition Pk
of X, Pk ∪{{x̄}} is not an adequate (k+1)-partition of X ∪{x̄}
• is 1-interior
is 2-interior
k-interior
Introdu tion
⊲ Interiority
Separability index
Convexity riterion
Auto-interiority
A general notion of interiority 6th Winter S hool on Te hnology Assessment � 12
entity x̄ is k-interior with respe t to X if for every k-partition Pk
of X, Pk ∪{{x̄}} is not an adequate (k+1)-partition of X ∪{x̄}
• is 1-interior
is 2-interior
k-interior
Introdu tion
⊲ Interiority
Separability index
Convexity riterion
Auto-interiority
A general notion of interiority 6th Winter S hool on Te hnology Assessment � 12
entity x̄ is k-interior with respe t to X if for every k-partition Pk
of X, Pk ∪{{x̄}} is not an adequate (k+1)-partition of X ∪{x̄}
• is 1-interior
• is 2-interior
k-interior
Introdu tion
⊲ Interiority
Separability index
Convexity riterion
Auto-interiority
A general notion of interiority 6th Winter S hool on Te hnology Assessment � 13
entity x̄ is k-interior with respe t to X if for every k-partition Pk
of X, Pk ∪{{x̄}} is not an adequate (k+1)-partition of X ∪{x̄}
is 1-interior but not 2-interior
k-interior
Introdu tion
⊲ Interiority
Separability index
Convexity riterion
Auto-interiority
A general notion of interiority 6th Winter S hool on Te hnology Assessment � 13
entity x̄ is k-interior with respe t to X if for every k-partition Pk
of X, Pk ∪{{x̄}} is not an adequate (k+1)-partition of X ∪{x̄}
is 1-interior but not 2-interior
k-interior
Introdu tion
⊲ Interiority
Separability index
Convexity riterion
Auto-interiority
A general notion of interiority 6th Winter S hool on Te hnology Assessment � 13
entity x̄ is k-interior with respe t to X if for every k-partition Pk
of X, Pk ∪{{x̄}} is not an adequate (k+1)-partition of X ∪{x̄}
is 1-interior but not 2-interior
k-interior
Introdu tion
⊲ Interiority
Separability index
Convexity riterion
Auto-interiority
A general notion of interiority 6th Winter S hool on Te hnology Assessment � 13
entity x̄ is k-interior with respe t to X if for every k-partition Pk
of X, Pk ∪{{x̄}} is not an adequate (k+1)-partition of X ∪{x̄}
• is 1-interior
but not 2-interior
k-interior
Introdu tion
⊲ Interiority
Separability index
Convexity riterion
Auto-interiority
A general notion of interiority 6th Winter S hool on Te hnology Assessment � 13
entity x̄ is k-interior with respe t to X if for every k-partition Pk
of X, Pk ∪{{x̄}} is not an adequate (k+1)-partition of X ∪{x̄}
• is 1-interior
but not 2-interior
k-interior
Introdu tion
⊲ Interiority
Separability index
Convexity riterion
Auto-interiority
A general notion of interiority 6th Winter S hool on Te hnology Assessment � 13
entity x̄ is k-interior with respe t to X if for every k-partition Pk
of X, Pk ∪{{x̄}} is not an adequate (k+1)-partition of X ∪{x̄}
• is 1-interior but not 2-interior
interiority degree
Introdu tion
⊲ Interiority
Separability index
Convexity riterion
Auto-interiority
A general notion of interiority 6th Winter S hool on Te hnology Assessment � 14
the interiority degree of is
the largest su h that is -interior
if is not interior
� diameter i�
� single linkage i�
,
where is a non trivial bipartition of
diameter and single linkage i�
interiority degree
Introdu tion
⊲ Interiority
Separability index
Convexity riterion
Auto-interiority
A general notion of interiority 6th Winter S hool on Te hnology Assessment � 14
the interiority degree of x̄ is
dx̄ = the largest k su h that x̄ is k-interior
if is not interior
� diameter i�
� single linkage i�
,
where is a non trivial bipartition of
diameter and single linkage i�
interiority degree
Introdu tion
⊲ Interiority
Separability index
Convexity riterion
Auto-interiority
A general notion of interiority 6th Winter S hool on Te hnology Assessment � 14
the interiority degree of x̄ is
dx̄ = the largest k su h that x̄ is k-interior
dx̄ = 0 if x̄ is not interior
� diameter i�
� single linkage i�
,
where is a non trivial bipartition of
diameter and single linkage i�
interiority degree
Introdu tion
⊲ Interiority
Separability index
Convexity riterion
Auto-interiority
A general notion of interiority 6th Winter S hool on Te hnology Assessment � 14
the interiority degree of x̄ is
dx̄ = the largest k su h that x̄ is k-interior
dx̄ = 0 if x̄ is not interior
� dx̄ = 0
� diameter i�
� single linkage i�
,
where is a non trivial bipartition of
diameter and single linkage i�
interiority degree
Introdu tion
⊲ Interiority
Separability index
Convexity riterion
Auto-interiority
A general notion of interiority 6th Winter S hool on Te hnology Assessment � 14
the interiority degree of x̄ is
dx̄ = the largest k su h that x̄ is k-interior
dx̄ = 0 if x̄ is not interior
� dx̄ = 0
� diameter i�
� single linkage i�
,
where is a non trivial bipartition of
diameter and single linkage i�
interiority degree
Introdu tion
⊲ Interiority
Separability index
Convexity riterion
Auto-interiority
A general notion of interiority 6th Winter S hool on Te hnology Assessment � 14
the interiority degree of x̄ is
dx̄ = the largest k su h that x̄ is k-interior
dx̄ = 0 if x̄ is not interior
� dx̄ = 0
� diameter i� minx∈X
dis(x̄, x) ≥ maxx,x′∈X
dis(x, x′)
� single linkage i�
,
where is a non trivial bipartition of
diameter and single linkage i�
interiority degree
Introdu tion
⊲ Interiority
Separability index
Convexity riterion
Auto-interiority
A general notion of interiority 6th Winter S hool on Te hnology Assessment � 14
the interiority degree of x̄ is
dx̄ = the largest k su h that x̄ is k-interior
dx̄ = 0 if x̄ is not interior
� dx̄ = 0
� diameter i� minx∈X
dis(x̄, x) ≥ maxx,x′∈X
dis(x, x′)
� single linkage i�
,
where is a non trivial bipartition of
diameter and single linkage i�
interiority degree
Introdu tion
⊲ Interiority
Separability index
Convexity riterion
Auto-interiority
A general notion of interiority 6th Winter S hool on Te hnology Assessment � 14
the interiority degree of x̄ is
dx̄ = the largest k su h that x̄ is k-interior
dx̄ = 0 if x̄ is not interior
� dx̄ = 0
� diameter i� minx∈X
dis(x̄, x) ≥ maxx,x′∈X
dis(x, x′)
� single linkage i�
minx∈X
dis(x̄, x) ≥ max{X′,X′′}
minx′∈X′,x′′∈X′′
dis(x′, x′′),
where {X ′, X ′′} is a non trivial bipartition of X
diameter and single linkage i�
interiority degree
Introdu tion
⊲ Interiority
Separability index
Convexity riterion
Auto-interiority
A general notion of interiority 6th Winter S hool on Te hnology Assessment � 14
the interiority degree of x̄ is
dx̄ = the largest k su h that x̄ is k-interior
dx̄ = 0 if x̄ is not interior
� dx̄ = 0
� diameter i� minx∈X
dis(x̄, x) ≥ maxx,x′∈X
dis(x, x′)
� single linkage i�
minx∈X
dis(x̄, x) ≥ max{X′,X′′}
minx′∈X′,x′′∈X′′
dis(x′, x′′),
where {X ′, X ′′} is a non trivial bipartition of X
� dx̄ = |X| − 1 diameter and single linkage i�
interiority degree
Introdu tion
⊲ Interiority
Separability index
Convexity riterion
Auto-interiority
A general notion of interiority 6th Winter S hool on Te hnology Assessment � 14
the interiority degree of x̄ is
dx̄ = the largest k su h that x̄ is k-interior
dx̄ = 0 if x̄ is not interior
� dx̄ = 0
� diameter i� minx∈X
dis(x̄, x) ≥ maxx,x′∈X
dis(x, x′)
� single linkage i�
minx∈X
dis(x̄, x) ≥ max{X′,X′′}
minx′∈X′,x′′∈X′′
dis(x′, x′′),
where {X ′, X ′′} is a non trivial bipartition of X
� dx̄ = |X| − 1 diameter and single linkage i�
minx∈X
dis(x̄, x) < minx,x′∈X
dis(x, x′)
this ombinatorial on ept of interiority...
Introdu tion
⊲ Interiority
Separability index
Convexity riterion
Auto-interiority
A general notion of interiority 6th Winter S hool on Te hnology Assessment � 15
de�nes, for ea h aggregation riterion, a parti ular partition of
the entity spa e into a �nite number of iso-interiority regions
this ombinatorial on ept of interiority...
Introdu tion
⊲ Interiority
Separability index
Convexity riterion
Auto-interiority
A general notion of interiority 6th Winter S hool on Te hnology Assessment � 15
de�nes, for ea h aggregation riterion, a parti ular partition of
the entity spa e into a �nite number of iso-interiority regions
this ombinatorial on ept of interiority...
Introdu tion
⊲ Interiority
Separability index
Convexity riterion
Auto-interiority
A general notion of interiority 6th Winter S hool on Te hnology Assessment � 15
de�nes, for ea h aggregation riterion, a parti ular partition of
the entity spa e into a �nite number of iso-interiority regions
d : E \X → {0, 1, . . . , |X| − 1}
this ombinatorial on ept of interiority...
Introdu tion
⊲ Interiority
Separability index
Convexity riterion
Auto-interiority
A general notion of interiority 6th Winter S hool on Te hnology Assessment � 15
de�nes, for ea h aggregation riterion, a parti ular partition of
the entity spa e into a �nite number of iso-interiority regions
d : E \X → {0, 1, . . . , |X| − 1}
separability index
Introdu tion
Interiority
⊲
Separability
index
Convexity riterion
Auto-interiority
A general notion of interiority 6th Winter S hool on Te hnology Assessment � 16
interiority of
dis riminate sep. index bla k/(bla k+red)
separability index
Introdu tion
Interiority
⊲
Separability
index
Convexity riterion
Auto-interiority
A general notion of interiority 6th Winter S hool on Te hnology Assessment � 16
DX̄ = {d(x̄) : x̄ ∈ X̄ = {•}} interiority of X̄
dis riminate sep. index bla k/(bla k+red)
separability index
Introdu tion
Interiority
⊲
Separability
index
Convexity riterion
Auto-interiority
A general notion of interiority 6th Winter S hool on Te hnology Assessment � 16
DX̄ = {d(x̄) : x̄ ∈ X̄ = {•}} interiority of X̄
DX̄ dis riminate sep. index DSI ∈ [0, 1]
dis riminate sep. index bla k/(bla k+red)
separability index
Introdu tion
Interiority
⊲
Separability
index
Convexity riterion
Auto-interiority
A general notion of interiority 6th Winter S hool on Te hnology Assessment � 16
DX̄ = {d(x̄) : x̄ ∈ X̄ = {•}} interiority of X̄
DX̄ dis riminate sep. index DSI ∈ [0, 1]
DSI of m points with interiority degree n = DSI of n points
with interiority degree m
dis riminate sep. index bla k/(bla k+red)
separability index
Introdu tion
Interiority
⊲
Separability
index
Convexity riterion
Auto-interiority
A general notion of interiority 6th Winter S hool on Te hnology Assessment � 16
DX̄ = {d(x̄) : x̄ ∈ X̄ = {•}} interiority of X̄
DX̄ dis riminate sep. index DSI ∈ [0, 1]
dis riminate sep. index bla k/(bla k+red)
separability index
Introdu tion
Interiority
⊲
Separability
index
Convexity riterion
Auto-interiority
A general notion of interiority 6th Winter S hool on Te hnology Assessment � 16
DX̄ = {d(x̄) : x̄ ∈ X̄ = {•}} interiority of X̄
DX̄ dis riminate sep. index DSI ∈ [0, 1]
dis riminate sep. index bla k/(bla k+red)
separability index
Introdu tion
Interiority
⊲
Separability
index
Convexity riterion
Auto-interiority
A general notion of interiority 6th Winter S hool on Te hnology Assessment � 16
DX̄ = {d(x̄) : x̄ ∈ X̄ = {•}} interiority of X̄
DX̄ dis riminate sep. index DSI = bla k/(bla k+red)
DSI = bla k/(bla k+red)
Introdu tion
Interiority
⊲
Separability
index
Convexity riterion
Auto-interiority
A general notion of interiority 6th Winter S hool on Te hnology Assessment � 17
DSI = bla k/(bla k+red)
Introdu tion
Interiority
⊲
Separability
index
Convexity riterion
Auto-interiority
A general notion of interiority 6th Winter S hool on Te hnology Assessment � 17
DSI = bla k/(bla k+red)
Introdu tion
Interiority
⊲
Separability
index
Convexity riterion
Auto-interiority
A general notion of interiority 6th Winter S hool on Te hnology Assessment � 17
DSI = bla k/(bla k+red)
Introdu tion
Interiority
⊲
Separability
index
Convexity riterion
Auto-interiority
A general notion of interiority 6th Winter S hool on Te hnology Assessment � 18
DSI = bla k/(bla k+red)
Introdu tion
Interiority
⊲
Separability
index
Convexity riterion
Auto-interiority
A general notion of interiority 6th Winter S hool on Te hnology Assessment � 18
DSI = bla k/(bla k+red)
Introdu tion
Interiority
⊲
Separability
index
Convexity riterion
Auto-interiority
A general notion of interiority 6th Winter S hool on Te hnology Assessment � 18
onvexity riterion
Introdu tion
Interiority
Separability index
⊲
Convexity
riterion
Auto-interiority
A general notion of interiority 6th Winter S hool on Te hnology Assessment � 19
a -partition is adequate if it in ludes singletons that are
not in the interior of the onvex hull of the remaining points
onvexity riterion
Introdu tion
Interiority
Separability index
⊲
Convexity
riterion
Auto-interiority
A general notion of interiority 6th Winter S hool on Te hnology Assessment � 19
a k-partition is adequate if it in ludes k − 1 singletons that are
not in the interior of the onvex hull of the remaining points
onvexity riterion
Introdu tion
Interiority
Separability index
⊲
Convexity
riterion
Auto-interiority
A general notion of interiority 6th Winter S hool on Te hnology Assessment � 19
a k-partition is adequate if it in ludes k − 1 singletons that are
not in the interior of the onvex hull of the remaining points
onvexity riterion
Introdu tion
Interiority
Separability index
⊲
Convexity
riterion
Auto-interiority
A general notion of interiority 6th Winter S hool on Te hnology Assessment � 19
a k-partition is adequate if it in ludes k − 1 singletons that are
not in the interior of the onvex hull of the remaining points
onvexity riterion
Introdu tion
Interiority
Separability index
⊲
Convexity
riterion
Auto-interiority
A general notion of interiority 6th Winter S hool on Te hnology Assessment � 19
a k-partition is adequate if it in ludes k − 1 singletons that are
not in the interior of the onvex hull of the remaining points
not adequate
onvexity riterion
Introdu tion
Interiority
Separability index
⊲
Convexity
riterion
Auto-interiority
A general notion of interiority 6th Winter S hool on Te hnology Assessment � 19
a k-partition is adequate if it in ludes k − 1 singletons that are
not in the interior of the onvex hull of the remaining points
not adequate
onvexity riterion
Introdu tion
Interiority
Separability index
⊲
Convexity
riterion
Auto-interiority
A general notion of interiority 6th Winter S hool on Te hnology Assessment � 19
a k-partition is adequate if it in ludes k − 1 singletons that are
not in the interior of the onvex hull of the remaining points
onvexity riterion
Introdu tion
Interiority
Separability index
⊲
Convexity
riterion
Auto-interiority
A general notion of interiority 6th Winter S hool on Te hnology Assessment � 19
a k-partition is adequate if it in ludes k − 1 singletons that are
not in the interior of the onvex hull of the remaining points
adequate
onvexity riterion
Introdu tion
Interiority
Separability index
⊲
Convexity
riterion
Auto-interiority
A general notion of interiority 6th Winter S hool on Te hnology Assessment � 19
a k-partition is adequate if it in ludes k − 1 singletons that are
not in the interior of the onvex hull of the remaining points
adequate
onvexity riterion
Introdu tion
Interiority
Separability index
⊲
Convexity
riterion
Auto-interiority
A general notion of interiority 6th Winter S hool on Te hnology Assessment � 19
a k-partition is adequate if it in ludes k − 1 singletons that are
not in the interior of the onvex hull of the remaining points
onvexity riterion
Introdu tion
Interiority
Separability index
⊲
Convexity
riterion
Auto-interiority
A general notion of interiority 6th Winter S hool on Te hnology Assessment � 19
a k-partition is adequate if it in ludes k − 1 singletons that are
not in the interior of the onvex hull of the remaining points
adequate
onvexity riterion
Introdu tion
Interiority
Separability index
⊲
Convexity
riterion
Auto-interiority
A general notion of interiority 6th Winter S hool on Te hnology Assessment � 19
a k-partition is adequate if it in ludes k − 1 singletons that are
not in the interior of the onvex hull of the remaining points
adequate
onvexity riterion
Introdu tion
Interiority
Separability index
⊲
Convexity
riterion
Auto-interiority
A general notion of interiority 6th Winter S hool on Te hnology Assessment � 20
x̄ is k-interior if it is inside the onvex hull of any |X| − k + 1
points of X
- (halfspa e, lo ation, Tukey) depth
onvexity riterion
Introdu tion
Interiority
Separability index
⊲
Convexity
riterion
Auto-interiority
A general notion of interiority 6th Winter S hool on Te hnology Assessment � 20
x̄ is k-interior if it is inside the onvex hull of any |X| − k + 1
points of X
- (halfspa e, lo ation, Tukey) depth
onvexity riterion
Introdu tion
Interiority
Separability index
⊲
Convexity
riterion
Auto-interiority
A general notion of interiority 6th Winter S hool on Te hnology Assessment � 20
x̄ is k-interior if it is inside the onvex hull of any |X| − k + 1
points of X
- (halfspa e, lo ation, Tukey) depth
onvexity riterion
Introdu tion
Interiority
Separability index
⊲
Convexity
riterion
Auto-interiority
A general notion of interiority 6th Winter S hool on Te hnology Assessment � 20
x̄ is k-interior if it is inside the onvex hull of any |X| − k + 1
points of X
• is 1-interior
- (halfspa e, lo ation, Tukey) depth
onvexity riterion
Introdu tion
Interiority
Separability index
⊲
Convexity
riterion
Auto-interiority
A general notion of interiority 6th Winter S hool on Te hnology Assessment � 20
x̄ is k-interior if it is inside the onvex hull of any |X| − k + 1
points of X
• is 1-interior, and 2-interior
- (halfspa e, lo ation, Tukey) depth
onvexity riterion
Introdu tion
Interiority
Separability index
⊲
Convexity
riterion
Auto-interiority
A general notion of interiority 6th Winter S hool on Te hnology Assessment � 20
x̄ is k-interior if it is inside the onvex hull of any |X| − k + 1
points of X
• is 1-interior, and 2-interior
d(•) is the minimum number of points to be removed from
X su h that • is outside onv. hull of the remaining points
- (halfspa e, lo ation, Tukey) depth
onvexity riterion
Introdu tion
Interiority
Separability index
⊲
Convexity
riterion
Auto-interiority
A general notion of interiority 6th Winter S hool on Te hnology Assessment � 20
x̄ is k-interior if it is inside the onvex hull of any |X| − k + 1
points of X
• is 1-interior, and 2-interior
d(•) is the minimum number of points to be removed from
X su h that • is outside onv. hull of the remaining points
d(•) = 2
- (halfspa e, lo ation, Tukey) depth
onvexity riterion
Introdu tion
Interiority
Separability index
⊲
Convexity
riterion
Auto-interiority
A general notion of interiority 6th Winter S hool on Te hnology Assessment � 20
x̄ is k-interior if it is inside the onvex hull of any |X| − k + 1
points of X
• is 1-interior, and 2-interior
d(•) is the minimum number of points to be removed from
X su h that • is outside onv. hull of the remaining points
d(•) = 2
d( ) - (halfspa e, lo ation, Tukey) depth
iso-interiority regions
Introdu tion
Interiority
Separability index
⊲
Convexity
riterion
Auto-interiority
A general notion of interiority 6th Winter S hool on Te hnology Assessment � 21
iso-interiority regions
Introdu tion
Interiority
Separability index
⊲
Convexity
riterion
Auto-interiority
A general notion of interiority 6th Winter S hool on Te hnology Assessment � 21
depth →֒ auto-interiority
Introdu tion
Interiority
Separability index
Convexity riterion
⊲ Auto-interiority
A general notion of interiority 6th Winter S hool on Te hnology Assessment � 22
for , is the minimum number of points to be removed
from su h that is outside the onv.hull of the remaining
points
i� is a vertex of the onv.hull
depth of
should distinguishes between on�gurations on entrated in the
"interior� of the onv.hull, and those whi h o ur mainly on the
"margins�...
This is relevant as it would allow dete tion the pattern of the
loud of points issue 2
depth →֒ auto-interiority
Introdu tion
Interiority
Separability index
Convexity riterion
⊲ Auto-interiority
A general notion of interiority 6th Winter S hool on Te hnology Assessment � 22
for x ∈ X, d(x) is the minimum number of points to be removed
from X su h that x is outside the onv.hull of the remaining
points
i� is a vertex of the onv.hull
depth of
should distinguishes between on�gurations on entrated in the
"interior� of the onv.hull, and those whi h o ur mainly on the
"margins�...
This is relevant as it would allow dete tion the pattern of the
loud of points issue 2
depth →֒ auto-interiority
Introdu tion
Interiority
Separability index
Convexity riterion
⊲ Auto-interiority
A general notion of interiority 6th Winter S hool on Te hnology Assessment � 22
for x ∈ X, d(x) is the minimum number of points to be removed
from X su h that x is outside the onv.hull of the remaining
points
d(x) = 1 i� x is a vertex of the onv.hull(X)
depth of
should distinguishes between on�gurations on entrated in the
"interior� of the onv.hull, and those whi h o ur mainly on the
"margins�...
This is relevant as it would allow dete tion the pattern of the
loud of points issue 2
depth →֒ auto-interiority
Introdu tion
Interiority
Separability index
Convexity riterion
⊲ Auto-interiority
A general notion of interiority 6th Winter S hool on Te hnology Assessment � 22
for x ∈ X, d(x) is the minimum number of points to be removed
from X su h that x is outside the onv.hull of the remaining
points
d(x) = 1 i� x is a vertex of the onv.hull(X)
DX = {d(x) : x ∈ X} depth of X
should distinguishes between on�gurations on entrated in the
"interior� of the onv.hull, and those whi h o ur mainly on the
"margins�...
This is relevant as it would allow dete tion the pattern of the
loud of points issue 2
depth →֒ auto-interiority
Introdu tion
Interiority
Separability index
Convexity riterion
⊲ Auto-interiority
A general notion of interiority 6th Winter S hool on Te hnology Assessment � 22
for x ∈ X, d(x) is the minimum number of points to be removed
from X su h that x is outside the onv.hull of the remaining
points
d(x) = 1 i� x is a vertex of the onv.hull(X)
DX = {d(x) : x ∈ X} depth of X
should distinguishes between on�gurations on entrated in the
"interior� of the onv.hull, and those whi h o ur mainly on the
"margins�...
This is relevant as it would allow dete tion the pattern of the
loud of points issue 2
depth →֒ auto-interiority
Introdu tion
Interiority
Separability index
Convexity riterion
⊲ Auto-interiority
A general notion of interiority 6th Winter S hool on Te hnology Assessment � 22
for x ∈ X, d(x) is the minimum number of points to be removed
from X su h that x is outside the onv.hull of the remaining
points
d(x) = 1 i� x is a vertex of the onv.hull(X)
DX = {d(x) : x ∈ X} depth of X
should distinguishes between on�gurations on entrated in the
"interior� of the onv.hull, and those whi h o ur mainly on the
"margins�...
This is relevant as it would allow dete tion the pattern of the
loud of points
issue 2
depth →֒ auto-interiority
Introdu tion
Interiority
Separability index
Convexity riterion
⊲ Auto-interiority
A general notion of interiority 6th Winter S hool on Te hnology Assessment � 22
for x ∈ X, d(x) is the minimum number of points to be removed
from X su h that x is outside the onv.hull of the remaining
points
d(x) = 1 i� x is a vertex of the onv.hull(X)
DX = {d(x) : x ∈ X} depth of X
should distinguishes between on�gurations on entrated in the
"interior� of the onv.hull, and those whi h o ur mainly on the
"margins�...
This is relevant as it would allow dete tion the pattern of the
loud of points →֒ issue 2
if X is uniformly distributed on an n-ball
Introdu tion
Interiority
Separability index
Convexity riterion
⊲ Auto-interiority
A general notion of interiority 6th Winter S hool on Te hnology Assessment � 23
Result holds if is uniformly distributed on the losed region of
delimited by an ellipsoid
if X is uniformly distributed on an n-ball
Introdu tion
Interiority
Separability index
Convexity riterion
⊲ Auto-interiority
A general notion of interiority 6th Winter S hool on Te hnology Assessment � 23
d(x)
|X|
Result holds if is uniformly distributed on the losed region of
delimited by an ellipsoid
if X is uniformly distributed on an n-ball
Introdu tion
Interiority
Separability index
Convexity riterion
⊲ Auto-interiority
A general notion of interiority 6th Winter S hool on Te hnology Assessment � 23
d(x)
|X|
vol( )
vol(Bn)= Vr(x)
Result holds if is uniformly distributed on the losed region of
delimited by an ellipsoid
if X is uniformly distributed on an n-ball
Introdu tion
Interiority
Separability index
Convexity riterion
⊲ Auto-interiority
A general notion of interiority 6th Winter S hool on Te hnology Assessment � 23
d(x)
|X|
vol( )
vol(Bn)= Vr(x)
E[Vr(x)] =
∫
Bn
Vr(x)
vol(Bn)dx
Result holds if is uniformly distributed on the losed region of
delimited by an ellipsoid
if X is uniformly distributed on an n-ball
Introdu tion
Interiority
Separability index
Convexity riterion
⊲ Auto-interiority
A general notion of interiority 6th Winter S hool on Te hnology Assessment � 23
d(x)
|X|
vol( )
vol(Bn)= Vr(x)
E[Vr(x)] =
∫
Bn
Vr(x)
vol(Bn)dx =
1
2n+1
Result holds if is uniformly distributed on the losed region of
delimited by an ellipsoid
if X is uniformly distributed on an n-ball
Introdu tion
Interiority
Separability index
Convexity riterion
⊲ Auto-interiority
A general notion of interiority 6th Winter S hool on Te hnology Assessment � 23
d(x)
|X|
vol( )
vol(Bn)= Vr(x)
E[Vr(x)] =
∫
Bn
Vr(x)
vol(Bn)dx =
1
2n+1
Result holds if X is uniformly distributed on the losed region of
Rn
delimited by an ellipsoid
if X is uniformly distributed on an n-ball
Introdu tion
Interiority
Separability index
Convexity riterion
⊲ Auto-interiority
A general notion of interiority 6th Winter S hool on Te hnology Assessment � 24
E[Vr(x)] =1
2n+1
if X is uniformly distributed on an n-ball
Introdu tion
Interiority
Separability index
Convexity riterion
⊲ Auto-interiority
A general notion of interiority 6th Winter S hool on Te hnology Assessment � 24
E[Vr(x)] =1
2n+1=⇒
1
|X|
∑x
Vr(x) ≈1
2n+1
if X is uniformly distributed on an n-ball
Introdu tion
Interiority
Separability index
Convexity riterion
⊲ Auto-interiority
A general notion of interiority 6th Winter S hool on Te hnology Assessment � 25
(a) (b) (c)
normal uniform ldi
0 1874 0 1247 0 0582
even for small , , for uniformly
distributed
if X is uniformly distributed on an n-ball
Introdu tion
Interiority
Separability index
Convexity riterion
⊲ Auto-interiority
A general notion of interiority 6th Winter S hool on Te hnology Assessment � 25
(a) (b) (c)
normal uniform ldi
1
|X|
∑x Vr(x) 0·1874 0·1247 0·0582
even for small , , for uniformly
distributed
if X is uniformly distributed on an n-ball
Introdu tion
Interiority
Separability index
Convexity riterion
⊲ Auto-interiority
A general notion of interiority 6th Winter S hool on Te hnology Assessment � 25
(a) (b) (c)
normal uniform ldi
1
|X|
∑x Vr(x) 0·1874 0·1247 0·0582
E[Vr(x)] =1
2n+1=
1
23= 0.125
even for small , , for uniformly
distributed
if X is uniformly distributed on an n-ball
Introdu tion
Interiority
Separability index
Convexity riterion
⊲ Auto-interiority
A general notion of interiority 6th Winter S hool on Te hnology Assessment � 25
(a) (b) (c)
normal uniform ldi
1
|X|
∑x Vr(x) 0·1874 0·1247 0·0582
E[Vr(x)] =1
2n+1=
1
23= 0.125
even for small |X|, 1
|X|
∑x Vr(x) ≈ E[Vr(x)] =
1
2n+1 , for uniformly
distributed X