Geometric Inference on Kernel Density Estimates
|
|
- Aubrey Shelton
- 5 years ago
- Views:
Transcription
1 Geometric Inference on Kernel Density Estimates Bei Wang 1 Jeff M. Phillips 2 Yan Zheng 2 1 Scientific Computing and Imaging Institute, University of Utah 2 School of Computing, University of Utah Feb 6, 2014
2 Take home message Geometric inference from a point cloud can be calculated by examining its kernel density estimate (KDE) of Gaussians. Such an inference is made possible with provable properties through the vehicle of kernel distance. Such an inference is robust to noise and scalable. We provide an algorithm to estimate the topology of kernel distance using weighted Vietoris-Rips complexes.
3 A bit more detail... Geometric inference using the kernel distance, in place of the distance to a measure [Chazal Cohen-Steiner Merigot 2011]. 1. [Robustness] Kernel distance is distance-like: 1-Lipschitz, 1-semiconcave, proper and stable. 2. [Scalability] Kernel distance has a small coreset, making efficient inference possible on 100 million points. 3. [Relation to KDE] Geometric inference based on kernel distance works naturally via superlevel sets of KDE: sublevel sets of the kernel distance are superlevel sets of KDE. 4. [Algorithm] to approximate the sublevel set filtration of kernel distance from a point cloud sample.
4 Why kernel distance? People love and are familiar with KDE, especially with Gaussian kernel Kernel distance provides a proper way to relate KDE with properties that are crucial for geometric inference We could approximate the topology of kernel distance via point cloud samples
5 Background on kernels, KDE and kernel distance
6 Kernels A kernel is a similarity measure, more similar points have higher value, K : R d R d R + We focus on the Gaussian kernel (positive definite):. K(p, x) = σ 2 exp( p x 2 /2σ 2 ) x x x Gaussian G Triangle T Ball B
7 Kernel density estimate (KDE) A kernel density estimate represents a continuous distribution function over R d for point set P R d : kde P (x) = 1 P K(p, x) p P More generally, it can be applied to any measure µ (on R d ) as kde µ (x) = K(p, x)µ(p)dp p R d
8 Kernel distance For two point sets P and Q, define similarity κ(p, Q) = 1 1 K(p, q) P Q If Q = {x}, κ(p, x) = kde P (x). p P q Q The kernel distance (a metric between P and Q): D K (P, Q) = κ(p, P ) + κ(q, Q) 2κ(P, Q) Self similarity minus cross similarity... [Phillips, Venkatasubramanian 2011]
9 Kernel distance For two point sets P and Q, define similarity κ(p, Q) = 1 1 K(p, q) P Q If Q = {x}, κ(p, x) = kde P (x). p P q Q The kernel distance (a metric between P and Q): D K (P, Q) = κ(p, P ) + κ(q, Q) 2κ(P, Q) Self similarity minus cross similarity... [Phillips, Venkatasubramanian 2011]
10 Kernel distance For two point sets P and Q, define similarity κ(p, Q) = 1 1 K(p, q) P Q If Q = {x}, κ(p, x) = kde P (x). p P q Q The kernel distance (a metric between P and Q): D K (P, Q) = κ(p, P ) + κ(q, Q) 2κ(P, Q) Self similarity minus cross similarity... [Phillips, Venkatasubramanian 2011]
11 Kernel distance For D K (µ, ν) between two measures µ and ν, define similarity κ(µ, ν) = K(p, q)µ(p)µ(q)dpdq p R d q R d The kernel distance (a metric between µ and ν): D K (µ, ν) = κ(µ, µ) + κ(ν, ν) 2κ(µ, ν) If ν = unit Dirac mass at x, κ(µ, x) = kde µ (x), D K (µ, x) = κ(µ, µ) + κ(x, x) 2κ(µ, x) = c µ 2kde µ (x) Kernel distance (current distance or maximum mean discrepancy) is a metric, if the kernel K is characteristic (a slight restriction of being positive definite, e.g. Gaussian and Laplace kernels).
12 Geometric inference and distance to a measure: a review [Lieutier 2004] [Chazal, Cohen-Steiner, Lieutier 2009] [Merigot, Ovsjanikov, Guibas 2009] [Chazal, Cohen-Steiner, Merigot 2010] [Biau, Chazal, Cohen-Steiner, Devroye and Rodriguez 2011] [Chazal Cohen-Steiner Merigot 2011]...
13 Geometric inference Given: An unknown object (e.g. a compact set) S R d A finite point cloud P R d that comes from S under some process Aim: Recover topological and geometric properties of S from P, e.g. # of components, dimension, curvature... e.g. preserve homeomorphism, homotopy type, or homology of S from P.
14 Distance function based geometric inference Reconstructs an approximation of S by offsets from P. [Chazal, Cohen-Steiner, Lieutier 2009] Distance function: f P (x) = inf y P x y Offset: (P ) r = f 1 P ([0, r]) Hausdorff distance: d H (S, P ) := f S f P = inf x R d f S (x) f P (x)
15 Distance function based geometric inference: the intuition [Hausdorff stability w.r.t. distance functions] If d H (S, P ) is small, thus f S and f P are close, and subsequently, S, (S) r and (P ) r carry the same topology for an appropriate scale r. Theorem (Reconstruction from f P ) Let S, P R d be compact sets such that reach(s) > R and ε := d H (S, P ) R/7. Then S and (P ) r are homotopy equivalent (and even isotopic) if 4ε r R 3ε. [Chazal Cohen-Steiner Lieutier 2009] [Chazal Cohen-Steiner Merigot 2011] R ensures topological properties of S and (S) r are the same; ε ensures (S) r and (P ) r are close, ε density of the sample.
16 Distance function based geometric inference Not robust to outliers. [Chazal Cohen-Steiner Merigot 2011] If S = S x and f S (x) > R, then f S f S > R : offset-based inference methods fail...
17 Distance(-like) function that is robust to noise... Desirable properties for f S to be useful in geometric inference: (F1) f S is 1-Lipschitz: for all x, y R d, f S (x) f S (y) x y. (F2) f 2 S is 1-semiconcave: x Rd (f S (x)) 2 x 2 is concave. (F1) ensures that f S is differentiable almost everywhere and the medial axis of S has zero d-volume; (F2) is crucial, e.g. in proving the existence of the flow of the gradient of the distance function for topological inference.
18 Distance to a measure [Chazal Cohen-Steiner Merigot 2011] Intuition: W 2 distance to m 0 fraction of the space. µ: probability measure on R d m 0 > 0: a parameter smaller than the total mass of µ The distance to a measure d CCM µ,m 0 : R n R +, x R d, ( 1 m0 ) 1/2 d CCM µ,m 0 (x) = (δ µ,m (x)) 2 dm m 0 m=0 where δ µ,m (x) = inf { r > 0 : µ( B r (x)) m }. ( ) Wasserstein-2 distance W 2 (µ, ν) = inf π Π(µ,ν) R d R d x y 2 1/2 dπ(x, y)
19 Distance to a measure d CCM µ,m 0 is distance-like (D1) 1-Lipschitz (D2) 1-semiconcave (D3) [Stability] For probability measures µ and ν on R d and m 0 > 0, then d CCM µ,m 0 d CCM ν,m 0 1 m0 W 2 (µ, ν). (D4) Proper (for Groves Isotopy Lemma). [Chazal Cohen-Steiner Merigot 2011]
20 Our Main Results
21 Our results Similar properties hold for the kernel distance defined as d K µ (x) = D K (µ, x) = κ(µ, µ) + κ(x, x) 2κ(µ, x) = c 2 µ 2kde µ (x) Specifically, the following properties of d K µ reconstruction properties of d CCM µ,m 0. allow it to inherit the (K1) d K µ is 1-Lipschitz. (K2) (d K µ ) 2 is 1-semiconvave: the map x (d K µ (x)) 2 x 2 is concave. (K3) d K µ is proper. (K4) [Stability] d K µ d K ν D K (µ, ν). For the point cloud setting, d K P (x) = D K (P, x) = κ(p, P ) + κ(x, x) 2κ(P, x) = c 2 P 2kde P (x)
22 Properness of d K µ A continuous map f : X Y between two topological spaces is proper if and only if the inverse image of every compact subset in Y is compact in X. Lemma (K3) d K µ is proper (when its range is restricted to be less than c µ ). Corollary The superlevel sets of kde µ for all ranges whose lower bound a > 0 are compact.
23 Reconstruction theory for d K µ Theorem (Isotopy lemma on d K µ ) Let r 1 < r 2 be two positive numbers such that d K µ has no critical points in (d K µ ) [r 1,r 2 ]. Then all the sublevel sets (d K µ ) r := {x R d d K µ (x) r} are isotopic for r [r 1, r 2 ]. Theorem (Reconstruction on d K µ, simple version) Let d K µ and d K ν be two kernel distance functions such that d K µ d K ν ε. Suppose reach(d K µ ) R. Then r [4ε, R 3ε], and η (0, R), the sublevel sets (d K µ ) η and (d K ν ) r are homotopy equivalent for ε R/9.
24 Stability properties for d K µ : Basic result For two probability measures µ and ν on R d, Lemma (K4) d K µ d K ν D K (µ, ν). Define µ = p p µ(p)dp. We have lim σ D K(µ, ν) = µ ν and µ ν W 2 (µ, ν). Theorem lim σ D K (µ, ν) W 2 (µ, ν). Open problem: remove need for limit in σ.
25 Stability properties for d K µ : Comparing D K to W 2 Lemma (One-sided bound) There is no Lipschitz constant γ s.t. for any µ and ν, we have W 2 (µ, ν) γd K (µ, ν). Intuition, W 2 can grow arbitrarily big, there are some measures from which our bound is tighter. Lemma (Special case) Consider µ and ν where ν is represented by a Dirac mass at a point x R d. Then d K µ (x) = D K (µ, ν) W 2 (µ, ν) for any σ > 0, where the equality only holds when µ is also a Dirac mass at x, and the difference decreases for any other x as σ increases.
26 Stability properties for d K µ : Stability with Respect to σ Theorem For any x, d K µ (x) is l-lipschitz with respect to σ, for l = 18/e 3 + 8/e + 2 < 6. σ geometric notion of an outlier parameter
27 Algorithm: Approximate the persistence diagram of sublevel sets filtration of kernel distance using weighted Rips filtration
28 Power distance construction [Buchet, Chazal, Oudot and Sheehy 2013] Metric space (X, d X (, )), a set P X and a function w : P R, the (general) power distance f associated with (P, w) is f(x) = min (d X(p, x) 2 + w(p) 2 ). p P Persistence diagram of w can be approximated by weight Rips filtration based on (P, w) and f.
29 Power distance construction [Buchet, Chazal, Oudot and Sheehy 2013] Sublevel set of f, f 1 ((, α]) is the union of balls centered at points p P with radius r p (α) = α 2 w(p) 2 for each p.
30 Power distance construction [Buchet, Chazal, Oudot and Sheehy 2013] Sublevel set of f, f 1 ((, α]) is the union of balls centered at points p P with radius r p (α) = α 2 w(p) 2 for each p. Weighted Čech complex C α (P, w) for parameter α is the union of simplices s such that p s B(p, r p(α)) 0.
31 Power distance construction [Buchet, Chazal, Oudot and Sheehy 2013] Sublevel set of f, f 1 ((, α]) is the union of balls centered at points p P with radius r p (α) = α 2 w(p) 2 for each p. Weighted Čech complex C α (P, w) for parameter α is the union of simplices s such that p s B(p, r p(α)) 0. Weighted Rips complex R α (P, w) for parameter α is the maximal complex whose 1-skeleton is the same as C α (P, w).
32 Power distance construction [Buchet, Chazal, Oudot and Sheehy 2013] Sublevel set of f, f 1 ((, α]) is the union of balls centered at points p P with radius r p (α) = α 2 w(p) 2 for each p. Weighted Čech complex C α (P, w) for parameter α is the union of simplices s such that p s B(p, r p(α)) 0. Weighted Rips complex R α (P, w) for parameter α is the maximal complex whose 1-skeleton is the same as C α (P, w). Weighted Rips filtration: {R α (P, w)}.
33 Power distance using d K µ A power distance using d K µ for a measure µ is defined with a point set P R d and a metric d(, ) on R d, ( f P (µ, x) = d(p, x) 2 + d K µ (p) 2). min p P
34 Power distance using d K µ A power distance using d K µ for a measure µ is defined with a point set P R d and a metric d(, ) on R d, ( f P (µ, x) = d(p, x) 2 + d K µ (p) 2). min p P We consider d(p, x) := D K (p, x). Let p + = arg max q R d κ(µ, q) and P + = P p +. We have 1 2 d K µ (x) f k P + (µ, x) 14d K µ (x).
35 Power distance using d K µ A power distance using d K µ for a measure µ is defined with a point set P R d and a metric d(, ) on R d, ( f P (µ, x) = d(p, x) 2 + d K µ (p) 2). min p P We consider d(p, x) := D K (p, x). Let p + = arg max q R d κ(µ, q) and P + = P p +. We have 1 2 d K µ (x) f k P + (µ, x) 14d K µ (x). However, constructing p + exactly seems quite difficult. We use its approximation ˆp + s.t. D K (P, ˆp + ) (1 + δ)d K (P, p + ) (obtained through an algorithm that is polynomial in n and 1/δ under reasonable conditions). Define ˆP + = P ˆp +.
36 Constructive topological estimation using d K µ Theorem The weighted Rips filtration {R α ( ˆP +, d K µ P )} can be used to approximate the persistence diagram of d K µ P such that d ln B (Dgm(dK µ P ), Dgm({R α ( ˆP +, d K µ P )})) ln(16). Proof based on the power distance mechanisms [Buchet, Chazal, Oudot and Sheehy 2013], ɛ-interleaving of persistence modules. Open problems: Let d(p, x) = x p Calculate ˆp + without restrictions, efficiently (current works, but messy) Tighten bounds
37 Comments on relating d K µ with f S D_K(x,0), sigma=2 3 x x /2 D_K(x,0), sigma= D_K(x,0), sigma=1/ Figure: Showing that x 0 /2 D K (x, 0) x 0, where the second inequality holds for x 3σ. The kernel distance D K (x, 0) is shown for σ = {1/2, 1, 2} in purple, blue, and red, respectively.
38 Algorithmic and Approximation Observations: Advantages
39 Advantages of the kernel distance summary (I) Small coreset representation for sparse representation and efficient, scalable computation. (II) Its inference is easily interpretable and computable through the superlevel sets of a KDE. (III) It is Lipschitz with respect to the outlier parameter σ when the input x is fixed. (IV) As σ, the kernel distance is bounded by the Wasserstein distance: lim σ D K (µ, ν) W 2 (µ, ν).
40 Small coreset There exists a small ɛ-coreset Q P s.t. d K P dk Q ε and kde P kde Q ε with probability at least 1 δ. Size O(((1/ε) log(1/εδ)) 2d/(d+2) ) [Phillips 2013]. The same holds under a random sample of size O((1/ε 2 )(d + log(1/δ))) [Joshi Kommaraju Phillips 2011]. Operate with P = 100,000,000 [Zheng Jestes Phillips Li 2013]. Stability of persistence diagram is preserved: d B (Dgm(kde P ), Dgm(kde Q )) ε.
41 Small coreset There exists a small ɛ-coreset Q P s.t. d K P dk Q ε and kde P kde Q ε with probability at least 1 δ. Size O(((1/ε) log(1/εδ)) 2d/(d+2) ) [Phillips 2013]. The same holds under a random sample of size O((1/ε 2 )(d + log(1/δ))) [Joshi Kommaraju Phillips 2011]. Operate with P = 100,000,000 [Zheng Jestes Phillips Li 2013]. Stability of persistence diagram is preserved: d B (Dgm(kde P ), Dgm(kde Q )) ε.
42 Geometric inference with KDE Recall d K P c (x) = 2 P 2kde P (x) where c 2 P is a constant that depends only on P. Perform geometric inference on noisy P by considering the super-level sets of kde P, Key: {x R d kde P (x) τ} d K P ( ) is monotonic with kde P ( ); as d K P (x) gets smaller, kde P (x) gets larger. A clean and natural interpretation of the reconstruction problem through the well-studied lens of KDE. Geometric inference with sublevel sets of d K P (superlevel sets of kde P ).
43 Experiments An example with 25% of P as noise, σ = 0.05
44 Experiments An example with 25% of P as noise, σ = 0.003
45 Experiments An example with 25% of P as noise, σ = 0.001
46 Comparison with distance to a measure (a) (b) Figure: Sublevel sets of kernel distance (a), and distance to a measure (b) on a data set with two circles with different scales and densities. These plots fix a level set γ and vary (a) σ for d K µ and (b) m 0 for d CCM The parameters are chosen to provide similar appearing images. µ,m 0.
47 Future directions
48 Beyond Gaussian kernels More general theory for KDE with systematic understanding of family of kernels: distance to a measure (KNN kernel), kernel distance (a larger class of kernels, e.g. Gaussian, Laplace; triangle kernel may work OK in practice with less perfect properties).
49 Alternative KDE Laplace kernel K(p, x) = exp( 2 x y /σ)
50 Alternative KDE Triangle kernel: K(x, p) = max { } 0, 1 p x σ=0.05
51 Alternative KDE Epanechnikov kernel: (reconstruction) } K(x, p) = max {0, 1 p x 2 (σ=0.05) 2
52 Alternative KDE Ball kernel: K(x, p) = { 1 if p x σ = otherwise. α shape can be viewed as using the ball kernel with σ = α and r = 1/n.
53 Alternative KDEs
54 Multi-dim / scale space persistence? Parameter selection? Two parameters r (isolevel) and σ (outlier/bandwidth) that control the scale. Figure: Sublevel sets for the kernel distance. Left: fix σ, vary r. Right: fix r, vary σ. The values of σ and r are chosen to make the plots similar.
55 Acknowlegement The authors thank D. Sheehy, F. Chazal and the rest of the Geometrica group at INRIA-Saclay for enlightening discussions on geometric and topological reconstruction. We also thank D. Sheehy for personal communications regarding the power distance constructions.
56 References Paper available at:
Some Topics in Computational Topology. Yusu Wang. AMS Short Course 2014
Some Topics in Computational Topology Yusu Wang AMS Short Course 2014 Introduction Much recent developments in computational topology Both in theory and in their applications E.g, the theory of persistence
More informationSome Topics in Computational Topology
Some Topics in Computational Topology Yusu Wang Ohio State University AMS Short Course 2014 Introduction Much recent developments in computational topology Both in theory and in their applications E.g,
More informationGeometric Inference for Probability distributions
Geometric Inference for Probability distributions F. Chazal 1 D. Cohen-Steiner 2 Q. Mérigot 2 1 Geometrica, INRIA Saclay, 2 Geometrica, INRIA Sophia-Antipolis 2009 June 1 Motivation What is the (relevant)
More informationGeometric Inference for Probability distributions
Geometric Inference for Probability distributions F. Chazal 1 D. Cohen-Steiner 2 Q. Mérigot 2 1 Geometrica, INRIA Saclay, 2 Geometrica, INRIA Sophia-Antipolis 2009 July 8 F. Chazal, D. Cohen-Steiner, Q.
More informationA sampling theory for compact sets
ENS Lyon January 2010 A sampling theory for compact sets F. Chazal Geometrica Group INRIA Saclay To download these slides: http://geometrica.saclay.inria.fr/team/fred.chazal/teaching/distancefunctions.pdf
More informationGeometric inference for measures based on distance functions The DTM-signature for a geometric comparison of metric-measure spaces from samples
Distance to Geometric inference for measures based on distance functions The - for a geometric comparison of metric-measure spaces from samples the Ohio State University wan.252@osu.edu 1/31 Geometric
More informationWITNESSED K-DISTANCE
WITNESSED K-DISTANCE LEONIDAS GUIBAS, DMITRIY MOROZOV, AND QUENTIN MÉRIGOT Abstract. Distance function to a compact set plays a central role in several areas of computational geometry. Methods that rely
More informationGeometric Inference for Probability Measures
Geometric Inference for Probability Measures Frédéric CHAZAL David COHEN-STEINER Quentin MÉRIGOT June 21, 2010 Abstract Data often comes in the form of a point cloud sampled from an unknown compact subset
More informationMetric Thickenings of Euclidean Submanifolds
Metric Thickenings of Euclidean Submanifolds Advisor: Dr. Henry Adams Committe: Dr. Chris Peterson, Dr. Daniel Cooley Joshua Mirth Masters Thesis Defense October 3, 2017 Introduction Motivation Can we
More informationStability of boundary measures
Stability of boundary measures F. Chazal D. Cohen-Steiner Q. Mérigot INRIA Saclay - Ile de France LIX, January 2008 Point cloud geometry Given a set of points sampled near an unknown shape, can we infer
More informationGeometric Inference on Kernel Density Estimates
1 Geometric Inference on Kernel Density Estimates Jeff M. Phillis jeff@cs.utah.edu University of Utah Bei Wang beiwang@sci.utah.edu University of Utah Yan Zheng yanzheng@cs.utah.edu University of Utah
More informationDeclutter and Resample: Towards parameter free denoising
Declutter and Resample: Towards parameter free denoising Mickaël Buchet, Tamal K. Dey, Jiayuan Wang, Yusu Wang arxiv:1511.05479v2 [cs.cg] 27 Mar 2017 Abstract In many data analysis applications the following
More informationWitnessed k-distance
Witnessed k-distance Leonidas J. Guibas, Quentin Mérigot, Dmitriy Morozov To cite this version: Leonidas J. Guibas, Quentin Mérigot, Dmitriy Morozov. Witnessed k-distance. Discrete and Computational Geometry,
More informationHomology through Interleaving
Notes by Tamal K. Dey, OSU 91 Homology through Interleaving 32 Concept of interleaving A discrete set P R k is assumed to be a sample of a set X R k, if it lies near to it which we can quantify with the
More informationParameter-free Topology Inference and Sparsification for Data on Manifolds
Parameter-free Topology Inference and Sparsification for Data on Manifolds Tamal K. Dey Zhe Dong Yusu Wang Abstract In topology inference from data, current approaches face two major problems. One concerns
More informationGraph Induced Complex: A Data Sparsifier for Homology Inference
Graph Induced Complex: A Data Sparsifier for Homology Inference Tamal K. Dey Fengtao Fan Yusu Wang Department of Computer Science and Engineering The Ohio State University October, 2013 Space, Sample and
More informationPersistent Homology of Data, Groups, Function Spaces, and Landscapes.
Persistent Homology of Data, Groups, Function Spaces, and Landscapes. Shmuel Weinberger Department of Mathematics University of Chicago William Benter Lecture City University of Hong Kong Liu Bie Ju Centre
More informationStatistics and Topological Data Analysis
Geometry Understanding in Higher Dimensions CHAIRE D INFORMATIQUE ET SCIENCES NUMÉRIQUES Collège de France - June 2017 Statistics and Topological Data Analysis Bertrand MICHEL Ecole Centrale de Nantes
More informationSubsampling in Smoothed Range Spaces
Subsampling in Smoothed Range Spaces Jeff M. Phillips and Yan Zheng University of Utah Abstract. We consider smoothed versions of geometric range spaces, so an element of the ground set (e.g. a point)
More informationInterleaving Distance between Merge Trees
Noname manuscript No. (will be inserted by the editor) Interleaving Distance between Merge Trees Dmitriy Morozov Kenes Beketayev Gunther H. Weber the date of receipt and acceptance should be inserted later
More informationScalar Field Analysis over Point Cloud Data
DOI 10.1007/s00454-011-9360-x Scalar Field Analysis over Point Cloud Data Frédéric Chazal Leonidas J. Guibas Steve Y. Oudot Primoz Skraba Received: 19 July 2010 / Revised: 13 April 2011 / Accepted: 18
More information(x k ) sequence in F, lim x k = x x F. If F : R n R is a function, level sets and sublevel sets of F are any sets of the form (respectively);
STABILITY OF EQUILIBRIA AND LIAPUNOV FUNCTIONS. By topological properties in general we mean qualitative geometric properties (of subsets of R n or of functions in R n ), that is, those that don t depend
More informationMetric shape comparison via multidimensional persistent homology
Metric shape comparison via multidimensional persistent homology Patrizio Frosini 1,2 1 Department of Mathematics, University of Bologna, Italy 2 ARCES - Vision Mathematics Group, University of Bologna,
More informationOn Inverse Problems in TDA
Abel Symposium Geiranger, June 2018 On Inverse Problems in TDA Steve Oudot joint work with E. Solomon (Brown Math.) arxiv: 1712.03630 Features Data Rn The preimage problem in the data Sciences feature
More information7. The Stability Theorem
Vin de Silva Pomona College CBMS Lecture Series, Macalester College, June 2017 The Persistence Stability Theorem Stability Theorem (Cohen-Steiner, Edelsbrunner, Harer 2007) The following transformation
More informationMinimax Rates. Homology Inference
Minimax Rates for Homology Inference Don Sheehy Joint work with Sivaraman Balakrishan, Alessandro Rinaldo, Aarti Singh, and Larry Wasserman Something like a joke. Something like a joke. What is topological
More informationThe Persistent Homology of Distance Functions under Random Projection
The Persistent Homology of Distance Functions under Random Projection Donald R. Sheehy University of Connecticut don.r.sheehy@gmail.com Abstract Given n points P in a Euclidean space, the Johnson-Linden-strauss
More informationTHESIS METRIC THICKENINGS OF EUCLIDEAN SUBMANIFOLDS. Submitted by. Joshua R. Mirth. Department of Mathematics
THESIS METRIC THICKENINGS OF EUCLIDEAN SUBMANIFOLDS Submitted by Joshua R. Mirth Department of Mathematics In partial fulfillment of the requirements For the Degree of Master of Science Colorado State
More informationGeodesic Delaunay Triangulations in Bounded Planar Domains
Geodesic Delaunay Triangulations in Bounded Planar Domains Leonidas J. Guibas Dept. of Computer Science Stanford University Stanford, CA guibas@cs.stanford.edu Jie Gao Dept. of Computer Science Stony Brook
More informationPersistent Homology: Course Plan
Persistent Homology: Course Plan Andrey Blinov 16 October 2017 Abstract This is a list of potential topics for each class of the prospective course, with references to the literature. Each class will have
More informationTOPOLOGY OF POINT CLOUD DATA
TOPOLOGY OF POINT CLOUD DATA CS 468 Lecture 8 3/3/4 Afra Zomorodian CS 468 Lecture 8 - Page 1 OVERVIEW Points Complexes Cěch Rips Alpha Filtrations Persistence Afra Zomorodian CS 468 Lecture 8 - Page 2
More informationGromov-Hausdorff Approximation of Filament Structure Using Reeb-type Graph
Noname manuscript No. (will be inserted by the editor) Gromov-Hausdorff Approximation of Filament Structure Using Reeb-type Graph Frédéric Chazal Ruqi Huang Jian Sun Received: date / Accepted: date Abstract
More information10 Typical compact sets
Tel Aviv University, 2013 Measure and category 83 10 Typical compact sets 10a Covering, packing, volume, and dimension... 83 10b A space of compact sets.............. 85 10c Dimensions of typical sets.............
More informationarxiv: v2 [math.ag] 24 Jun 2015
TRIANGULATIONS OF MONOTONE FAMILIES I: TWO-DIMENSIONAL FAMILIES arxiv:1402.0460v2 [math.ag] 24 Jun 2015 SAUGATA BASU, ANDREI GABRIELOV, AND NICOLAI VOROBJOV Abstract. Let K R n be a compact definable set
More informationCourse 212: Academic Year Section 1: Metric Spaces
Course 212: Academic Year 1991-2 Section 1: Metric Spaces D. R. Wilkins Contents 1 Metric Spaces 3 1.1 Distance Functions and Metric Spaces............. 3 1.2 Convergence and Continuity in Metric Spaces.........
More informationGromov-Hausdorff Approximation of Filament Structure Using Reeb-type Graph
Gromov-Hausdorff Approximation of Filament Structure Using Reeb-type Graph Frédéric Chazal and Ruqi Huang and Jian Sun July 26, 2014 Abstract In many real-world applications data appear to be sampled around
More informationPersistence based Clustering
July 9, 2009 TGDA Workshop Persistence based Clustering Primoz Skraba joint work with Frédéric Chazal, Steve Y. Oudot, Leonidas J. Guibas 1-1 Clustering Input samples 2-1 Clustering Input samples Important
More informationA Higher-Dimensional Homologically Persistent Skeleton
A Higher-Dimensional Homologically Persistent Skeleton Sara Kališnik Verovšek, Vitaliy Kurlin, Davorin Lešnik 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 Abstract A data set is often given as a point cloud,
More informationLearning from persistence diagrams
Learning from persistence diagrams Ulrich Bauer TUM July 7, 2015 ACAT final meeting IST Austria Joint work with: Jan Reininghaus, Stefan Huber, Roland Kwitt 1 / 28 ? 2 / 28 ? 2 / 28 3 / 28 3 / 28 3 / 28
More informationPersistent Local Homology: Theory, Applications, Algorithms
Persistent Local Homology: Theory, Applications, Algorithms Paul Bendich Duke University Feb. 4, 2014 Basic Goal: stratification learning Q: Given a point cloud sampled from a stratified space, how do
More informationg 2 (x) (1/3)M 1 = (1/3)(2/3)M.
COMPACTNESS If C R n is closed and bounded, then by B-W it is sequentially compact: any sequence of points in C has a subsequence converging to a point in C Conversely, any sequentially compact C R n is
More informationOptimal Reconstruction Might be Hard
Optimal Reconstruction Might be Hard Dominique Attali, André Lieutier To cite this version: Dominique Attali, André Lieutier. Optimal Reconstruction Might be Hard. Discrete and Computational Geometry,
More informationMultivariate Topological Data Analysis
Cleveland State University November 20, 2008, Duke University joint work with Gunnar Carlsson (Stanford), Peter Kim and Zhiming Luo (Guelph), and Moo Chung (Wisconsin Madison) Framework Ideal Truth (Parameter)
More information1 Lesson 1: Brunn Minkowski Inequality
1 Lesson 1: Brunn Minkowski Inequality A set A R n is called convex if (1 λ)x + λy A for any x, y A and any λ [0, 1]. The Minkowski sum of two sets A, B R n is defined by A + B := {a + b : a A, b B}. One
More informationTopology Guaranteeing Manifold Reconstruction using Distance Function to Noisy Data
Topology Guaranteeing Manifold Reconstruction using Distance Function to Noisy Data ABSTRACT Frédéric Chazal Université de Bourgogne Institute of Mathematics of Burgundy France fchazal@u-bourgogne.fr Given
More informationChapter 2 Linear Transformations
Chapter 2 Linear Transformations Linear Transformations Loosely speaking, a linear transformation is a function from one vector space to another that preserves the vector space operations. Let us be more
More informationReeb Graphs: Approximation and Persistence
Reeb Graphs: Approximation and Persistence Tamal K. Dey Yusu Wang Abstract Given a continuous function f : X IR on a topological space X, its level set f 1 (a) changes continuously as the real value a
More informationA Higher-Dimensional Homologically Persistent Skeleton
arxiv:1701.08395v5 [math.at] 10 Oct 2017 A Higher-Dimensional Homologically Persistent Skeleton Sara Kališnik Verovšek, Vitaliy Kurlin, Davorin Lešnik Abstract Real data is often given as a point cloud,
More informationLocal semiconvexity of Kantorovich potentials on non-compact manifolds
Local semiconvexity of Kantorovich potentials on non-compact manifolds Alessio Figalli, Nicola Gigli Abstract We prove that any Kantorovich potential for the cost function c = d / on a Riemannian manifold
More informationAnalysis of Scalar Fields over Point Cloud Data
Analysis of Scalar Fields over Point Cloud Data Frédéric Chazal Leonidas J. Guibas Steve Y. Oudot Primoz Skraba Abstract Given a real-valued function f defined over some metric space X, is it possible
More informationProximity of Persistence Modules and their Diagrams
Proximity of Persistence Modules and their Diagrams Frédéric Chazal David Cohen-Steiner Marc Glisse Leonidas J. Guibas Steve Y. Oudot November 28, 2008 Abstract Topological persistence has proven to be
More informationTopological properties of Z p and Q p and Euclidean models
Topological properties of Z p and Q p and Euclidean models Samuel Trautwein, Esther Röder, Giorgio Barozzi November 3, 20 Topology of Q p vs Topology of R Both R and Q p are normed fields and complete
More informationStability of Critical Points with Interval Persistence
Stability of Critical Points with Interval Persistence Tamal K. Dey Rephael Wenger Abstract Scalar functions defined on a topological space Ω are at the core of many applications such as shape matching,
More informationLECTURE 15: COMPLETENESS AND CONVEXITY
LECTURE 15: COMPLETENESS AND CONVEXITY 1. The Hopf-Rinow Theorem Recall that a Riemannian manifold (M, g) is called geodesically complete if the maximal defining interval of any geodesic is R. On the other
More informationLinear-Size Approximations to the Vietoris Rips Filtration
Linear-Size Approximations to the Vietoris Rips Filtration Donald R. Sheehy March 8, 2013 Abstract The Vietoris Rips filtration is a versatile tool in topological data analysis. It is a sequence of simplicial
More informationMinimal time mean field games
based on joint works with Samer Dweik and Filippo Santambrogio PGMO Days 2017 Session Mean Field Games and applications EDF Lab Paris-Saclay November 14th, 2017 LMO, Université Paris-Sud Université Paris-Saclay
More informationQUALITY AND EFFICIENCY IN KERNEL DENSITY ESTIMATES FOR LARGE DATA
QUALITY AND EFFICIENCY IN KERNEL DENSITY ESTIMATES FOR LARGE DATA Yan Zheng Jeffrey Jestes Jeff M. Philips Feifei Li University of Utah Salt Lake City Sigmod 2014 Snowbird Kernel Density Estimates Let
More information1 Spaces and operations Continuity and metric spaces Topological spaces Compactness... 3
Compact course notes Topology I Fall 2011 Professor: A. Penskoi transcribed by: J. Lazovskis Independent University of Moscow December 23, 2011 Contents 1 Spaces and operations 2 1.1 Continuity and metric
More informationTopological properties
CHAPTER 4 Topological properties 1. Connectedness Definitions and examples Basic properties Connected components Connected versus path connected, again 2. Compactness Definition and first examples Topological
More informationStratification Learning through Homology Inference
Stratification Learning through Homology Inference Paul Bendich Sayan Mukherjee Institute of Science and Technology Austria Department of Statistical Science Klosterneuburg, Austria Duke University, Durham
More informationII - REAL ANALYSIS. This property gives us a way to extend the notion of content to finite unions of rectangles: we define
1 Measures 1.1 Jordan content in R N II - REAL ANALYSIS Let I be an interval in R. Then its 1-content is defined as c 1 (I) := b a if I is bounded with endpoints a, b. If I is unbounded, we define c 1
More informationInference of curvature using tubular neighborhoods
Inference of curvature using tubular neighborhoods Frédéric Chazal, David Cohen-Steiner, André Lieutier Quentin Mérigot, Boris Thibert Introduction Geometric inference deals with the problem of recovering
More informationChapter 3: Baire category and open mapping theorems
MA3421 2016 17 Chapter 3: Baire category and open mapping theorems A number of the major results rely on completeness via the Baire category theorem. 3.1 The Baire category theorem 3.1.1 Definition. A
More informationTOPOLOGY OF POINT CLOUD DATA
TOPOLOGY OF POINT CLOUD DATA CS 468 Lecture 8 11/12/2 Afra Zomorodian CS 468 Lecture 8 - Page 1 PROJECTS Writeups: Introduction to Knot Theory (Giovanni De Santi) Mesh Processing with Morse Theory (Niloy
More informationAnalysis Finite and Infinite Sets The Real Numbers The Cantor Set
Analysis Finite and Infinite Sets Definition. An initial segment is {n N n n 0 }. Definition. A finite set can be put into one-to-one correspondence with an initial segment. The empty set is also considered
More informationA Topological Study of Functional Data and Fréchet Functions of Metric Measure Spaces
A Topological Study of Functional Data and Fréchet Functions of Metric Measure Spaces arxiv:1811.06613v1 [math.at] 15 Nov 2018 Haibin Hang, Facundo Mémoli, Washington Mio Department of Mathematics, Florida
More informationMath 5210, Definitions and Theorems on Metric Spaces
Math 5210, Definitions and Theorems on Metric Spaces Let (X, d) be a metric space. We will use the following definitions (see Rudin, chap 2, particularly 2.18) 1. Let p X and r R, r > 0, The ball of radius
More informationApproximating Cycles in a Shortest Basis of the First Homology Group from Point Data
Approximating Cycles in a Shortest Basis of the First Homology Group from Point Data Tamal K. Dey Jian Sun Yusu Wang Abstract Inference of topological and geometric attributes of a hidden manifold from
More informationCut Locus and Topology from Surface Point Data
Cut Locus and Topology from Surface Point Data Tamal K. Dey Kuiyu Li March 24, 2009 Abstract A cut locus of a point p in a compact Riemannian manifold M is defined as the set of points where minimizing
More informationThe Intersection of Statistics and Topology:
The Intersection of Statistics and Topology: Confidence Sets for Persistence Diagrams Brittany Terese Fasy joint work with S. Balakrishnan, F. Lecci, A. Rinaldo, A. Singh, L. Wasserman 3 June 2014 [FLRWBS]
More informationStability and computation of topological invariants of solids in R n
Stability and computation of topological invariants of solids in R n Frédéric Chazal André Lieutier Abstract In this work, one proves that under quite general assumptions one can deduce the topology of
More informationConvex Analysis and Economic Theory Winter 2018
Division of the Humanities and Social Sciences Ec 181 KC Border Convex Analysis and Economic Theory Winter 2018 Supplement A: Mathematical background A.1 Extended real numbers The extended real number
More informationFast and Accurate HARDI and its Application to Neurological Diagnosis
Fast and Accurate HARDI and its Application to Neurological Diagnosis Dr. Oleg Michailovich Department of Electrical and Computer Engineering University of Waterloo June 21, 2011 Outline 1 Diffusion imaging
More information9 Classification. 9.1 Linear Classifiers
9 Classification This topic returns to prediction. Unlike linear regression where we were predicting a numeric value, in this case we are predicting a class: winner or loser, yes or no, rich or poor, positive
More informationPersistence Theory: From Quiver Representations to Data Analysis
Persistence Theory: From Quiver Representations to Data Analysis Comments and Corrections Steve Oudot June 29, 2017 Comments page viii, bottom of page: the following names should be added to the acknowledgements:
More informationAn Introduction to Topological Data Analysis
An Introduction to Topological Data Analysis Elizabeth Munch Duke University :: Dept of Mathematics Tuesday, June 11, 2013 Elizabeth Munch (Duke) CompTop-SUNYIT Tuesday, June 11, 2013 1 / 43 There is no
More informationAnalysis III Theorems, Propositions & Lemmas... Oh My!
Analysis III Theorems, Propositions & Lemmas... Oh My! Rob Gibson October 25, 2010 Proposition 1. If x = (x 1, x 2,...), y = (y 1, y 2,...), then is a distance. ( d(x, y) = x k y k p Proposition 2. In
More informationLecture 9 Metric spaces. The contraction fixed point theorem. The implicit function theorem. The existence of solutions to differenti. equations.
Lecture 9 Metric spaces. The contraction fixed point theorem. The implicit function theorem. The existence of solutions to differential equations. 1 Metric spaces 2 Completeness and completion. 3 The contraction
More informationChapter 8. P-adic numbers. 8.1 Absolute values
Chapter 8 P-adic numbers Literature: N. Koblitz, p-adic Numbers, p-adic Analysis, and Zeta-Functions, 2nd edition, Graduate Texts in Mathematics 58, Springer Verlag 1984, corrected 2nd printing 1996, Chap.
More informationLecture Notes in Advanced Calculus 1 (80315) Raz Kupferman Institute of Mathematics The Hebrew University
Lecture Notes in Advanced Calculus 1 (80315) Raz Kupferman Institute of Mathematics The Hebrew University February 7, 2007 2 Contents 1 Metric Spaces 1 1.1 Basic definitions...........................
More informationHouston Journal of Mathematics c 2009 University of Houston Volume 35, No. 1, 2009
Houston Journal of Mathematics c 2009 University of Houston Volume 35, No. 1, 2009 ON THE GEOMETRY OF SPHERES WITH POSITIVE CURVATURE MENG WU AND YUNHUI WU Communicated by David Bao Abstract. For an n-dimensional
More informationarxiv: v2 [math.pr] 15 May 2016
MAXIMALLY PERSISTENT CYCLES IN RANDOM GEOMETRIC COMPLEXES OMER BOBROWSKI, MATTHEW KAHLE, AND PRIMOZ SKRABA arxiv:1509.04347v2 [math.pr] 15 May 2016 Abstract. We initiate the study of persistent homology
More informationMIT 9.520/6.860, Fall 2017 Statistical Learning Theory and Applications. Class 19: Data Representation by Design
MIT 9.520/6.860, Fall 2017 Statistical Learning Theory and Applications Class 19: Data Representation by Design What is data representation? Let X be a data-space X M (M) F (M) X A data representation
More informationOptimality Conditions for Nonsmooth Convex Optimization
Optimality Conditions for Nonsmooth Convex Optimization Sangkyun Lee Oct 22, 2014 Let us consider a convex function f : R n R, where R is the extended real field, R := R {, + }, which is proper (f never
More informationHomework in Topology, Spring 2009.
Homework in Topology, Spring 2009. Björn Gustafsson April 29, 2009 1 Generalities To pass the course you should hand in correct and well-written solutions of approximately 10-15 of the problems. For higher
More informationMetric Spaces and Topology
Chapter 2 Metric Spaces and Topology From an engineering perspective, the most important way to construct a topology on a set is to define the topology in terms of a metric on the set. This approach underlies
More informationHopf-Rinow and Hadamard Theorems
Summersemester 2015 University of Heidelberg Riemannian geometry seminar Hopf-Rinow and Hadamard Theorems by Sven Grützmacher supervised by: Dr. Gye-Seon Lee Prof. Dr. Anna Wienhard Contents Introduction..........................................
More informationDuality of multiparameter Hardy spaces H p on spaces of homogeneous type
Duality of multiparameter Hardy spaces H p on spaces of homogeneous type Yongsheng Han, Ji Li, and Guozhen Lu Department of Mathematics Vanderbilt University Nashville, TN Internet Analysis Seminar 2012
More informationOptimization and Optimal Control in Banach Spaces
Optimization and Optimal Control in Banach Spaces Bernhard Schmitzer October 19, 2017 1 Convex non-smooth optimization with proximal operators Remark 1.1 (Motivation). Convex optimization: easier to solve,
More informationThe Minimum Speed for a Blocking Problem on the Half Plane
The Minimum Speed for a Blocking Problem on the Half Plane Alberto Bressan and Tao Wang Department of Mathematics, Penn State University University Park, Pa 16802, USA e-mails: bressan@mathpsuedu, wang
More informationApproximating the Pathway Axis and the Persistence Diagram of a Collection of Balls in 3-Space
Approximating the Pathway Axis and the Persistence Diagram of a Collection of Balls in 3-Space Eitan Yaffe Dan Halperin School of Computer Science Tel-Aviv University, Israel {eitanyaf,danha}@post.tau.ac.il
More informationB. Appendix B. Topological vector spaces
B.1 B. Appendix B. Topological vector spaces B.1. Fréchet spaces. In this appendix we go through the definition of Fréchet spaces and their inductive limits, such as they are used for definitions of function
More informationTopological Simplification in 3-Dimensions
Topological Simplification in 3-Dimensions David Letscher Saint Louis University Computational & Algorithmic Topology Sydney Letscher (SLU) Topological Simplification CATS 2017 1 / 33 Motivation Original
More informationMATH 31BH Homework 1 Solutions
MATH 3BH Homework Solutions January 0, 04 Problem.5. (a) (x, y)-plane in R 3 is closed and not open. To see that this plane is not open, notice that any ball around the origin (0, 0, 0) will contain points
More informationarxiv: v1 [math.at] 27 Jul 2012
STATISTICAL TOPOLOGY USING PERSISTENCE LANDSCAPES PETER BUBENIK arxiv:1207.6437v1 [math.at] 27 Jul 2012 Abstract. We define a new descriptor for persistent homology, which we call the persistence landscape,
More informationTHE ASYMPTOTIC BEHAVIOUR OF HEEGAARD GENUS
THE ASYMPTOTIC BEHAVIOUR OF HEEGAARD GENUS MARC LACKENBY 1. Introduction Heegaard splittings have recently been shown to be related to a number of important conjectures in 3-manifold theory: the virtually
More informationPersistent homology and nonparametric regression
Cleveland State University March 10, 2009, BIRS: Data Analysis using Computational Topology and Geometric Statistics joint work with Gunnar Carlsson (Stanford), Moo Chung (Wisconsin Madison), Peter Kim
More informationMeasurable Choice Functions
(January 19, 2013) Measurable Choice Functions Paul Garrett garrett@math.umn.edu http://www.math.umn.edu/ garrett/ [This document is http://www.math.umn.edu/ garrett/m/fun/choice functions.pdf] This note
More informationCones of measures. Tatiana Toro. University of Washington. Quantitative and Computational Aspects of Metric Geometry
Cones of measures Tatiana Toro University of Washington Quantitative and Computational Aspects of Metric Geometry Based on joint work with C. Kenig and D. Preiss Tatiana Toro (University of Washington)
More informationarxiv: v5 [math.mg] 6 Sep 2018
PERSISTENT HOMOLOGY AND THE UPPER BOX DIMENSION BENJAMIN SCHWEINHART arxiv:1802.00533v5 [math.mg] 6 Sep 2018 Abstract. We introduce a fractal dimension for a metric space based on the persistent homology
More information