Geometric and Topological Data Analysis

12
Marcel Duchamp

Transcript of Geometric and Topological Data Analysis

Page 1: Geometric and Topological Data Analysis

Marcel Duchamp

Page 2: Geometric and Topological Data Analysis

3 Standard Stoppages

Page 3: Geometric and Topological Data Analysis

Network of Stoppages

Page 4: Geometric and Topological Data Analysis

Geometric and Topological Data Analysis

Geometric Data

Shape Analysis

Geometry Processing

Regression (space between data)

Clustering (meaning of data)

Geometry of Data

Persistent Homology (beyond linear structure)

(beyond simple connectivity)

Page 5: Geometric and Topological Data Analysis

Homology

Homology turns topological questions into algebraic questions.

Augment the data points with edges, triangles, tetrahedra, etc. !The kth boundary matrix ∂k maps k-simplices to the (k-1)-simplices in their boundary. !The kth homology group is the quotient ker ∂k / im ∂k+1. !Homology encodes connected components, holes, and voids.

Page 6: Geometric and Topological Data Analysis

Persistent Homologytime

H0

H1

Barcode

Persistence Diagram

Page 7: Geometric and Topological Data Analysis

Stability

Data set Persistence diagram

close Data set

close Persistence diagram

Page 8: Geometric and Topological Data Analysis

Stability

Data set Persistence diagramMetric on

sMetric on

sLipschitz

Page 9: Geometric and Topological Data Analysis

Computing Persistent HomologyInput: Boundary Matrix D Find V, R such that

D = RVV is upper-triangular R is “reduced” (i.e. no two columns have lowest nonzeros in the same row)

It’s just Gaussian elimination!

Output is a collection of pairs corresponding to the lowest nonzeros in R.

Page 10: Geometric and Topological Data Analysis

Nested DissectionA method for solving symmetric positive definite linear systems.

Ax = bIf A is n x n, consider the n vertex graph with an edge (i,j) for each nonzero entry A(i, j) of A. !Find a vertex separator S such that

- |S| = O(nβ) - each connected piece has at most cn vertices (for some c < 1). !

Repeat. Order the pivots going up from the leaves of the recursion.

The Punchline: Inverting A can be done in O(nβω) time.

Also works for computing ranks of singular, nonsymmetric matrices over finite fields.

Page 11: Geometric and Topological Data Analysis

Reasonable complexes have small separators.

The theory of geometric separators applies to graphs of nice meshes. !Separators on graphs can be “lifted” to separators on complexes. !Improves the asymptotic complexity of static homology. !Persistence?

Page 12: Geometric and Topological Data Analysis

Thanks.

Some open problems. !

How do we reconcile the filtration order and the nested dissection order?

!Is there a quotient version of

nested dissection? !

Is there a reasonable separator theory for filtrations?