seminar algorithms jorge corderokbuchin/.../slides/simplicialcomplexesforcloudsof… · 4 cech...

53
Simplicial complexes associated with cloud points Seminar algorithms Jorge Cordero Eindhoven University of Technology 8 May 2018

Upload: others

Post on 23-Oct-2020

1 views

Category:

Documents


0 download

TRANSCRIPT

  • Simplicial complexes associated with cloud pointsSeminar algorithms

    Jorge Cordero

    Eindhoven University of Technology

    8 May 2018

  • Outline

    1 Motivation

    2 Recap

    3 Nerves

    4 Čech complex

    5 Vietoris-Rips complex

    6 Delaunay complex

    7 Alpha complexes

  • Topological data analysis

    Data can be complex in terms of size or features.

    Sometimes, data has shape.

  • Topological data analysis

    Topological data analysis (TDA) help us to understand the structure(shape) of data.

    Application:

    Find coverage in networks of sensors

    Understand protein interactions

    Credit card fraud detection

  • Simplex

    σ is an n-simplex spanned by n + 1 affinely independent points in P ∈ Rn.

    A 0-simplex is a point

    A 1-simplex is a line

    A 2-simplex is a triangle

    A 3-simplex is a tetrahedron

  • Simplicial complex

    A simplicial complex K in Rn is a collection of simplices in Rn such that:

    Every face of a simplex of K is in K

    Every pair of distinct simplices of K has a disjoint interior

    The intersection of two distinct simplices of K is a face of each ofthem

  • ASC and Homotopy

    An abstract simplicial complex is a collection of finite non-empty subsetsof S.

    If A ⊆ S , each non-empty subset B ⊆ A is also in S .

    For our purposes, we can think of homotopy equivalent spaces (X ' Y) asspaces that can be deformed continuously one into another.

  • ASC and Homotopy

    An abstract simplicial complex is a collection of finite non-empty subsetsof S.

    If A ⊆ S , each non-empty subset B ⊆ A is also in S .

    For our purposes, we can think of homotopy equivalent spaces (X ' Y) asspaces that can be deformed continuously one into another.

  • Simplicial representation

    Our goal is to compute a simplicial representation of a set of points toapply topological tools for data analysis.

  • Nerves

    We can use the nerve of a finite collection of sets S to create an abstractsimplicial complex.

    The nerve of S consists of all non-empty subcollections whose sets have anon-empty common intersection,

    NrvS = {X ⊆ S |⋂

    X 6= ∅}.

  • Nerves

    We can use the nerve of a finite collection of sets S to create an abstractsimplicial complex.

    The nerve of S consists of all non-empty subcollections whose sets have anon-empty common intersection,

    NrvS = {X ⊆ S |⋂

    X 6= ∅}.

  • Nerves

    From⋂X 6= ∅ and Y ⊆ X =⇒

    ⋂Y 6= ∅, it follows that NrvS is always

    an abstract simplicial complex.

    We represent the nerve as an abstract simplicial complex,

    NrvS = {R,B,P,G , {R,B}, {B,P}, {P,G}, {G ,R}}.

  • Nerves

    The topological space of S = {R,G ,B,Y } is a disk with three holes.

    NrvS has the homotopy type of a sphere.

    Hence, the homotopy types for NrvS and |S | are different.

  • Nerves

    We want to compute a nerve that resembles the structure inherent to theset of points.

    Nerve theorem

    Let S be a finite collection of closed, convex sets in Euclidean space.Then, the nerve of S and the union of the sets in S have the samehomotopy type.

  • Nerves

    We want to compute a nerve that resembles the structure inherent to theset of points.

    Nerve theorem

    Let S be a finite collection of closed, convex sets in Euclidean space.Then, the nerve of S and the union of the sets in S have the samehomotopy type.

  • Closed ball

    A closed ball with center x and radius r is defined by,

    Bx(r) = x + rBd = {y ∈ Rd | ||y − x || ≤ r}

  • Čech complex

    Let’s construct a simplicial complex from a set of points.

  • Čech complex

    Let’s construct a simplicial complex from a set of points.

  • Čech complex

    Let’s construct a simplicial complex from a set of points.

  • Čech complex

    Let’s construct a simplicial complex from a set of points.

  • Čech complex

    Let S be a finite set of points in Rd .

    The Čech complex of S and r is given by,

    Č ech(r) = {σ ⊆ S |∩Bx(r) 6= ∅}

    The Čech complex is equivalent to the nerve of the collection of balls.

  • Čech complex

    Čech complexes for different values of r

  • Čech complex

    Čech complexes for different values of r

  • Čech complex

    Some properties of the Čech complex:

    When ri ≤ rj , Č ech(ri ) ⊆ Č ech(rj).

    Č ech(r) of a set S of points in Rd can always be represented as anabstract simplicial complex.

    Computing the complex takes an exponential time in the size of S .

  • Čech complex

    Some properties of the Čech complex:

    When ri ≤ rj , Č ech(ri ) ⊆ Č ech(rj).Č ech(r) of a set S of points in Rd can always be represented as anabstract simplicial complex.

    Computing the complex takes an exponential time in the size of S .

  • Čech complex

    Some properties of the Čech complex:

    When ri ≤ rj , Č ech(ri ) ⊆ Č ech(rj).Č ech(r) of a set S of points in Rd can always be represented as anabstract simplicial complex.

    Computing the complex takes an exponential time in the size of S .

  • Vietoris-Rips complex

    The Vietoris-Rips VR complex of S and r consists of all subsets ofdiameter at most 2r ,

    VR(r) = {σ ⊆ S |diamσ ≤ 2r}.

    The diameter of σ is the supremum of all pairwise distances betweenpoints in σ.

  • Comparing Čech and Vietoris-Rips complex

    Čech complex

    Vietoris-Rips complex

  • Vietoris-Rips

    Notice that Č ech(r) ⊆ VR(r) but Č ech(r) 6' VR(r).

    For appropriate values of ra and rb, we have VR(ra) ⊆ Č ech(rb).

    Vietoris-Rips Lemma

    Let S be a finite set of points in some Euclidean space and r ≥ 0. Itfollows that VR(r) ⊆ Č ech(

    √2r).

  • Vietoris-Rips

    Properties of VR complexes:

    VR complexes avoid the intersection test used in Čech.

    VR complexes also take exponential time to compute.

    VR complexes might not have a geometric representation in theunderlying space of S.

  • Vietoris-Rips

    Properties of VR complexes:

    VR complexes avoid the intersection test used in Čech.

    VR complexes also take exponential time to compute.

    VR complexes might not have a geometric representation in theunderlying space of S.

  • Vietoris-Rips

    Properties of VR complexes:

    VR complexes avoid the intersection test used in Čech.

    VR complexes also take exponential time to compute.

    VR complexes might not have a geometric representation in theunderlying space of S.

  • Voronoi diagram

    What if we want the simplicial complex to have a geometricrepresentation?.

    Delaunay complexes can be used for this task.

    To explore Delaunay complexes, we first introduce the Voronoi diagram.

  • Voronoi diagram

    Consider a finite set of points S = {v1, v2, ..., vn} in Rd :

    The Voronoi cell of vi ∈ S is the set of points in Rd closest to vi ,

    Vvi = {x ∈ Rd | ||x − vi || ≤ ||x − vj ||, ∀vj ∈ S}.

    Together, the Voronoi cells of all points vi cover the entire space.

  • Voronoi diagram

    For a set of points in a plane:

    Encode proximity information useful for answering point queries.

    Can be computed in time O(nlogn) using Fortune’s algorithm.

    It uses O(n) space.

  • Voronoi diagram

    For a set of points in a plane:

    Encode proximity information useful for answering point queries.

    Can be computed in time O(nlogn) using Fortune’s algorithm.

    It uses O(n) space.

  • Voronoi diagram

    For a set of points in a plane:

    Encode proximity information useful for answering point queries.

    Can be computed in time O(nlogn) using Fortune’s algorithm.

    It uses O(n) space.

  • Delaunay complex

    The Delaunay complex of a finite set of points S ⊆ Rd is equivalent to thenerve of the Voronoi diagram,

    Delaunay(S) = {σ ⊆ S |⋂ui∈σ

    Vui 6= ∅}.

  • Delaunay complex

    The Delaunay complex seems to always create triangles in R2.

    Is this always the case?

    No, unless the set of points is in general position.

  • Delaunay complex

    The Delaunay complex seems to always create triangles in R2.

    Is this always the case?

    No, unless the set of points is in general position.

  • Delaunay complex

    The Delaunay complex seems to always create triangles in R2.

    Is this always the case?

    No, unless the set of points is in general position.

  • Delaunay complex

    A set of points S is in general position if not d + 2 points lie on a common(d − 1)-sphere.

    For a finite set of points S ∈ Rd , assuming general position, the geometricrealization of a Delaunay complex fits in Rd .

  • Delaunay complex

    For a set of points in the plane:

    It can be computed from the Voronoi diagram.

    It takes expected O(nlogn) time.

    It uses O(n) space.

  • Alpha complexes

    Let S be a finite set of points in Rd and r ≥ 0.

    Let Ru(r) = Bu(r) ∩ Vu, with Vu being the Voronoi cell of u.

    The alpha complex is defined as

    Alpha(r) = {σ ⊆ S |⋂u∈σ

    Ru(r) 6= ∅}.

  • Delaunay filtration

    Different values of r to create different alpha complexes.

  • Delaunay filtration

    Eventually, we obtain the Delaunay complex.

  • Delaunay filtration

    The filtration of Km = Delaunay is represented as,

    ∅ = K0 ⊆ ... ⊆ Ki ⊆ ... ⊆ Km.

    Ki corresponds to the i-th alpha complex from the sequence of differentalpha complexes obtained by varying r .

  • Summary

    The nerve of a set of points S allows us to compute simplicialcomplexes.

    Čech complexes and Vietoris-Rips complexes allows us to constructabstract simplicial complexes from S .

    The Delaunay triangulation allows to create simplicial complexes withgeometric realizations.

    The alpha complex can be seen as a filtered version of the Delaunaycomplex.

  • Summary

    The nerve of a set of points S allows us to compute simplicialcomplexes.

    Čech complexes and Vietoris-Rips complexes allows us to constructabstract simplicial complexes from S .

    The Delaunay triangulation allows to create simplicial complexes withgeometric realizations.

    The alpha complex can be seen as a filtered version of the Delaunaycomplex.

  • Summary

    The nerve of a set of points S allows us to compute simplicialcomplexes.

    Čech complexes and Vietoris-Rips complexes allows us to constructabstract simplicial complexes from S .

    The Delaunay triangulation allows to create simplicial complexes withgeometric realizations.

    The alpha complex can be seen as a filtered version of the Delaunaycomplex.

  • Summary

    The nerve of a set of points S allows us to compute simplicialcomplexes.

    Čech complexes and Vietoris-Rips complexes allows us to constructabstract simplicial complexes from S .

    The Delaunay triangulation allows to create simplicial complexes withgeometric realizations.

    The alpha complex can be seen as a filtered version of the Delaunaycomplex.

  • Sources

    H. Edelsbrunner, J. L. Harer, Computational topology. Anintroduction. American Mathematical Society, Providence, RI, 2010.

    Mark de Berg, Otfried Cheong, Marc van Kreveld, and MarkOvermars. 2008. Computational Geometry: Algorithms andApplications (3rd ed.).

    Lecture notes and videos on Computational Topology. Lectures 11,12, and 13.

    Topological data analysis of genes

    Čech and Vietoris-Rips complex

    Persistent homology

    http://www.math.wsu.edu/faculty/bkrishna/FilesMath574/S16/LecNotes/welcome.htmlhttps://www.ayasdi.com/blog/topology/shape-genome-going-beyond-double-helix-topological-data-analysis/https://jeremykun.com/2015/08/06/cech-vietoris-rips-complex/https://www.youtube.com/watch?annotation_id=annotation_405791327&feature=iv&src_vid=h0bnG1Wavag&v=2PSqWBIrn90

  • Images

    https://mathematica.stackexchange.com/questions/61021/

    triangulated-mesh-from-voronoi-diagram

    https:

    //www.researchgate.net/figure/Delaunay-Triangulation_fig3_44897706

    http://www.pnas.org/content/pnas/113/26/7035/F10.large.jpg

    http://www.ugc.edu.hk/minisite/rgc_newsletter/rgcnews18/eng/05.htm

    http://www.pnas.org/content/pnas/113/26/7035/F10.large.jpg

    http://www.dei.unipd.it/~schenato/pics/SensorNetwork.jpg

    https://www.johndcook.com/persistent_homology.png

    https:

    //upload.wikimedia.org/wikipedia/commons/thumb/5/50/Simplicial_

    complex_example.svg/532px-Simplicial_complex_example.svg.png

    https://s3.amazonaws.com/cdn.ayasdi.com/wp-content/uploads/2017/05/

    09112616/srep43845-f2.jpg

    https://mathematica.stackexchange.com/questions/61021/triangulated-mesh-from-voronoi-diagramhttps://mathematica.stackexchange.com/questions/61021/triangulated-mesh-from-voronoi-diagramhttps://www.researchgate.net/figure/Delaunay-Triangulation_fig3_44897706https://www.researchgate.net/figure/Delaunay-Triangulation_fig3_44897706http://www.pnas.org/content/pnas/113/26/7035/F10.large.jpg http://www.ugc.edu.hk/minisite/rgc_newsletter/rgcnews18/eng/05.htmhttp://www.pnas.org/content/pnas/113/26/7035/F10.large.jpghttp://www.dei.unipd.it/~schenato/pics/SensorNetwork.jpghttps://www.johndcook.com/persistent_homology.pnghttps://upload.wikimedia.org/wikipedia/commons/thumb/5/50/Simplicial_complex_example.svg/532px-Simplicial_complex_example.svg.pnghttps://upload.wikimedia.org/wikipedia/commons/thumb/5/50/Simplicial_complex_example.svg/532px-Simplicial_complex_example.svg.pnghttps://upload.wikimedia.org/wikipedia/commons/thumb/5/50/Simplicial_complex_example.svg/532px-Simplicial_complex_example.svg.pnghttps://s3.amazonaws.com/cdn.ayasdi.com/wp-content/uploads/2017/05/09112616/srep43845-f2.jpghttps://s3.amazonaws.com/cdn.ayasdi.com/wp-content/uploads/2017/05/09112616/srep43845-f2.jpg

    MotivationRecapNervesCech complexVietoris-Rips complexDelaunay complexAlpha complexes