Data Analysis of Multi-level systems

Slides:



Advertisements
Similar presentations
Chapter 4 Euclidean Vector Spaces
Advertisements

3D Geometry for Computer Graphics
Nonlinear Dimension Reduction Presenter: Xingwei Yang The powerpoint is organized from: 1.Ronald R. Coifman et al. (Yale University) 2. Jieping Ye, (Arizona.
Mathematical Analysis of Complex Networks and Databases Philippe Blanchard Dima Volchenkov.
Non-linear Dimensionality Reduction CMPUT 466/551 Nilanjan Ray Prepared on materials from the book Non-linear dimensionality reduction By Lee and Verleysen,
. Markov Chains as a Learning Tool. 2 Weather: raining today40% rain tomorrow 60% no rain tomorrow not raining today20% rain tomorrow 80% no rain tomorrow.
1 Markov Chains (covered in Sections 1.1, 1.6, 6.3, and 9.4)
10/11/2001Random walks and spectral segmentation1 CSE 291 Fall 2001 Marina Meila and Jianbo Shi: Learning Segmentation by Random Walks/A Random Walks View.
Entropy Rates of a Stochastic Process
Ronald R. Coifman , Stéphane Lafon, 2006
Cosimo Stornaiolo INFN-Sezione di Napoli MG 12 Paris July 2009.
Prolog Algebra / Analysis vs Geometry Relativity → Riemannian Geometry Symmetry → Lie Derivatives → Lie Group → Lie Algebra Integration → Differential.
Computer Graphics Recitation 5.
General Relativity Physics Honours 2008 A/Prof. Geraint F. Lewis Rm 557, A29 Lecture Notes 2.
A Global Geometric Framework for Nonlinear Dimensionality Reduction Joshua B. Tenenbaum, Vin de Silva, John C. Langford Presented by Napat Triroj.
NonLinear Dimensionality Reduction or Unfolding Manifolds Tennenbaum|Silva|Langford [Isomap] Roweis|Saul [Locally Linear Embedding] Presented by Vikas.
STOCHASTIC GEOMETRY AND RANDOM GRAPHS FOR THE ANALYSIS AND DESIGN OF WIRELESS NETWORKS Haenggi et al EE 360 : 19 th February 2014.
Monte Carlo Methods in Partial Differential Equations.
Diffusion Maps and Spectral Clustering
Algorithms on negatively curved spaces James R. Lee University of Washington Robert Krauthgamer IBM Research (Almaden) TexPoint fonts used in EMF. Read.
Graphics CSE 581 – Interactive Computer Graphics Mathematics for Computer Graphics CSE 581 – Roger Crawfis (slides developed from Korea University slides)
A Study of The Applications of Matrices and R^(n) Projections By Corey Messonnier.
Hamdy N.Abd-ellah حمدي نور الدين عبد الله Department of Mathematics, Faculty of Science, Assiut University Assiut, Egypt جامعة أم القرى قسم الرياضيات.
Elementary Linear Algebra Anton & Rorres, 9th Edition
General (point-set) topology Jundong Liu Ohio Univ.
KINEMATICS of a ROLLING BALL Wayne Lawton Department of Mathematics National University of Singapore Lecture based on my student’s MSc.
Geometry of Shape Manifolds
Slides are modified from Lada Adamic
AGC DSP AGC DSP Professor A G Constantinides©1 Signal Spaces The purpose of this part of the course is to introduce the basic concepts behind generalised.
Information Geometry and Model Reduction Sorin Mitran 1 1 Department of Mathematics, University of North Carolina, Chapel Hill, NC, USA Reconstruction.
Math 285 Project Diffusion Maps Xiaoyan Chong Department of Mathematics and Statistics San Jose State University.
COMS Network Theory Week 5: October 6, 2010 Dragomir R. Radev Wednesdays, 6:10-8 PM 325 Pupin Terrace Fall 2010.
Geometric diffusions as a tool for harmonic analysis and structure definition of data By R. R. Coifman et al. The second-round discussion* on * The first-round.
1 Bhupendra Nath Tiwari IIT Kanpur in collaboration with T. Sarkar & G. Sengupta. Thermodynamic Geometry and BTZ black holes This talk is mainly based.
Graphics Graphics Korea University kucg.korea.ac.kr Mathematics for Computer Graphics 고려대학교 컴퓨터 그래픽스 연구실.
Random Walks for Data Analysis Dima Volchenkov (Bielefeld University) Discrete and Continuous Models in the Theory of Networks.
Random Walks and Diffusions on Networks and Databases Dimitri Volchenkov (Bielefeld University)
Introduction to Random Walks and Diffusions to Network and Databases: from Electric Networks to Urban Spatial Networks Dimitri Volchenkov (Bielefeld University.
Is it possible to geometrize infinite graphs?
Mathematical Analysis of Complex Networks and Databases
Spectral Methods for Dimensionality
Institutions do not die
Geometrize everything with Monge-Kantorovich?
Path-integral distance for the data analysis
Shan Lu, Jieqi Kang, Weibo Gong, Don Towsley UMASS Amherst
Random Walks for Data Analysis
Ca’ Foscari University of Venice;
Introduction The central problems of Linear Algebra are to study the properties of matrices and to investigate the solutions of systems of linear equations.
Introduction The central problems of Linear Algebra are to study the properties of matrices and to investigate the solutions of systems of linear equations.
Intrinsic Data Geometry from a Training Set
Industrial Engineering Dep
Structure creates a chance
Random remarks about random walks
Markov Chains Mixing Times Lecture 5
Unsupervised Riemannian Clustering of Probability Density Functions
Chapter 3 Formalism.
DTMC Applications Ranking Web Pages & Slotted ALOHA
Dipartimento di Ingegneria «Enzo Ferrari»,
Differential Manifolds and Tensors
GROUPS & THEIR REPRESENTATIONS: a card shuffling approach
Quantum One.
Goodfellow: Chapter 14 Autoencoders
An Introduction to Variational Methods for Graphical Models
1.3 Vector Equations.
Lecture 15: Least Square Regression Metric Embeddings
Linear Equations and Vectors
NonLinear Dimensionality Reduction or Unfolding Manifolds
Math review - scalars, vectors, and matrices
Shan Lu, Jieqi Kang, Weibo Gong, Don Towsley UMASS Amherst
Presentation transcript:

Data Analysis of Multi-level systems D. Volchenkov

Goals & Objectives To identify meaningful levels in multi-level systems; To understand the geometric structure associated to the multimodal data (i.e. how these levels are related to each other); To analyze the data geometrical manifold and its sensitivity to assorted data variations (evolution and control by Ricci-Hamilton flows). Processing data from other work packages.

We proceed in two steps: Step 1: “Probabilistic graph theory” Nodes, subgraphs (sets of nodes), graphs are described by probability distributions & characteristic times w.r.t. inhomogeneous Markov chains based on stochastic/fractional automorphisms (i.e., different definitions of a neighborhood); Step 2: “Geometrization of Data Manifolds” Establish geometric relations between those probability distributions whenever possible; 1. Coarse-graining/reduction/PCA for networks/databases → data analysis; 2. Transport optimization(Monge-Kontorovich type problems) → distances between distributions; 3. Network evolution & control via Ricci flows → sensitivity to assorted data variations;

Our heroes: Stochastic / fractional automorphisms Finite connected graphs; Finite distance matrices; Finite relational databases: the graph of attributes; the graph of tuples; GA (adjacency matrix of the graph)  P: [P,A]=0, P  Aut(G) Symmetry:

Our heroes: Stochastic / fractional automorphisms Finite connected graphs; Finite distance matrices; Finite relational databases: the graph of attributes; the graph of tuples; GA (adjacency matrix of the graph)  P: [P,A]=0, P  Aut(G) Symmetry: P =1, only trivial automorphisms! A permutation matrix is a particular case of stochastic matrix: Let us extend the notion of automorphisms onto the class of stochastic matrices.  T: [T, A]=0, Fractional automorphisms, or stochastic automorphisms

A variety of stochastic automorphisms at different scales Stochastic matrices: … …

A variety of stochastic automorphisms at different scales Neighborhoods & Scales: Stochastic matrices: Equiprobable paths … … … …

A variety of stochastic automorphisms at different scales Neighborhoods & Scales: Left eigenvectors (m=1) = Centrality measures: Stochastic matrices: Equiprobable paths The “stationary distribution” of the nearest neighbor RW … … … … … …

Relation to random walks Time is introduced as powers of transition matrices

Relation to random walks Time is introduced as powers of transition matrices

Relation to random walks Time is introduced as powers of transition matrices

Relation to random walks Time is introduced as powers of transition matrices

Relation to random walks Time is introduced as powers of transition matrices

Relation to random walks Time is introduced as powers of transition matrices

Relation to random walks Time is introduced as powers of transition matrices

Relation to random walks Time is introduced as powers of transition matrices

Relation to random walks Time is introduced as powers of transition matrices

Relation to random walks Time is introduced as powers of transition matrices

Relation to random walks Time is introduced as powers of transition matrices

Relation to random walks Time is introduced as powers of transition matrices

Relation to random walks Time is introduced as powers of transition matrices

Relation to random walks Time is introduced as powers of transition matrices

Relation to random walks Time is introduced as powers of transition matrices

Relation to random walks Time is introduced as powers of transition matrices

Relation to random walks Time is introduced as powers of transition matrices

Relation to random walks Time is introduced as powers of transition matrices

Relation to random walks Time is introduced as powers of transition matrices

Relation to random walks Time is introduced as powers of transition matrices Still far from stationary distribution! Stationary distribution is reached! Defect irrelevant. Low centrality (defect) repelling.

Research questions on stochastic automorphisms: 1. Generalized transition operator =? S i j 2. Properties of the Fourier transform of the transition operators: 3. Intuition about interscale Ricci flows on the data manifold: (approached faster)

Step 1: “Probabilistic graph theory” As soon as we define a neighborhood… Graph Subgraph (a subset of nodes) Node Time scale Tr T The probability that the RW stays at the initial node in 1 step. “Wave functions” (Slater determinants) of transients (traversing nodes and subgraphs within the characteristic scales) return the probability amplitudes whose modulus squared represent the probability density over the subgraphs. Probabilistic graph invariants = the t-steps recurrence probabilities quantifying the chance to return in t steps. … | det T | The probability that the RW revisits the initial node in N steps. Return times to the subgraphs within transients = 1/Pr{ … } Centrality measures (stationary distributions) Return times to a node Random target time Mixing times over subgraphs (times until the Markov chain is "close" to the steady state distribution)

As soon as we get probability distributions… Step 2: “Geometrization of Data Manifolds” As soon as we get probability distributions… Given T, L ≡ 1- T , the linear operators acting on distributions. The Green function is the natural way to find the relation between two distributions within the diffusion process Drazin’s generalized inverse:

As soon as we get probability distributions… Step 2: “Geometrization of Data Manifolds” As soon as we get probability distributions… Given T, L ≡ 1- T , the linear operators acting on distributions. The Green function is the natural way to find the relation between two distributions within the diffusion process Drazin’s generalized inverse: Given two distributions x,y over the set of nodes, we can define a scalar product, The (squared) norm of a vector and an angle The Euclidean distance:

Transport problems of the Monge-Kontorovich type Step 2: “Geometrization of Data Manifolds” As soon as we get probability distributions… Given T, L ≡ 1- T , the linear operators acting on distributions. The Green function is the natural way to find the relation between two distributions within the diffusion process Drazin’s generalized inverse: Given two distributions x,y over the set of nodes, we can define a scalar product, Transport problems of the Monge-Kontorovich type The (squared) norm of a vector and an angle The Euclidean distance: “First-passage transportation” from x to y x y W(x→y) W(y→x) ≠

Transport problems of the Monge-Kontorovich type Step 2: “Geometrization of Data Manifolds” As soon as we get probability distributions… Given T, L ≡ 1- T , the linear operators acting on distributions. The Green function is the natural way to find the relation between two distributions within the diffusion process Drazin’s generalized inverse: Given two distributions x,y over the set of nodes, we can define a scalar product, Transport problems of the Monge-Kontorovich type The (squared) norm of a vector and an angle The Euclidean distance: (Mean) first-passage time Commute time Electric potential Effective resistance distance Tax assessment land price in cities Musical diatonic scale degree … Musical tonality scale

As soon as we get probability distributions… Step 2: “Geometrization of Data Manifolds” As soon as we get probability distributions… Given T, L ≡ 1- T , the linear operators acting on distributions. The Green function is the natural way to find the relation between two distributions within the diffusion process Drazin’s generalized inverse: Given two distributions x,y over the set of nodes, we can define a scalar product, L◊ as the PCA kernel Low dimensional approximations; Graphic representations; Coarse-graining& reduction of networks; Data analysis; Data interpretation The (squared) norm of a vector and an angle The Euclidean distance:

Generalized Ricci curvature How fast the volume grows w.r.t. to geodesic one… Ricci curvature (Ollivier): On a metric space, for any two distinct points (x and y), the Ricci –Ollivier curvature along (x,y) is

Generalized Ricci curvature How fast the volume grows w.r.t. to geodesic one… Ricci curvature (Ollivier): On a metric space, for any two distinct points (x and y), the Ricci –Ollivier curvature along (x,y) is Similar concepts (“Ricci like curvatures”) w.r.t. stochastic automorphisms: Along two points: Trees are flat! At a point: Trees are flat!

An example of “curvature”: Tonality scales of Western music V.A. Mozart, Eine-Kleine-Nachtmusik R. Wagner, Das Rheingold (Entrance of Gods) Increase of harmonic interval/ first –passage time

Geometrization of undirected graphs Each node is characterized by a vector in RN-1 TxM RN-1 x ui uj p Riemannian metric: A node/entry dependent differential operator on the probabilistic manifold: For the group of translations, the shift operator is … and then define the Hessian matrix at each node It happens often that Q(x) forms a basis of orthogonal sections at x, g is a metric tensor. → the standard calculus of differential geometry…

Evolution of networks: Ricci-Hamilton flows We consider the metric tensor to be a function of a variable which is usually called "time”, then we obtain the geometric evolution equation (which preserves the volume of the metric): The Ricci flow tends to expand negatively curved regions of the manifold, and contract positively curved regions (“Mathew principle”) “Densification” of the network of “positive curvature” “Contraction” of a “probabilistic manifold” A “collapse” and decomposition of the network of “negative curvature”

Global connectedness: 1.How do we define a neighborhood? Concluding remarks Dealing with multi-level systems requires answering two basic questions: Global connectedness: 2. How do we connect neighborhoods? Local connectedness: 1.How do we define a neighborhood?

Global connectedness: Intra-scale Ricci flow → Concluding remarks Dealing with multi-level systems requires answering two basic questions: Global connectedness: 2. How do we connect neighborhoods? Intra-scale Ricci flow → t • Intra-scale Ricci curvature • Wasserstein type distances   Local connectedness: ← Inter-scale Ricci flow V grows differently over different subgraphs , so that 1.How do we define a neighborhood? Scales ℓ Vℓ grows differently at different ℓ, so Inter-scale Ricci curvature

t scales Concluding remarks   Real dynamical flows Dealing with multi-level systems requires answering two basic questions: Global connectedness: 2. How do we connect neighborhoods? Intra-scale Ricci flow → t • Intra-scale Ricci curvature • Wasserstein type distances   Local connectedness: ← Inter-scale Ricci flow 1.How do we define a neighborhood? scales Real dynamical flows Inter-scale Ricci curvature

Research questions on the Ricci flows : Given the inter-scale and intra-scale Ricci flows, is there a “scaling relations”, 2. Is it possible to define “universality classes” of different graphs/networks w.r.t a? 3. Scaffolding of real dynamical flows close to fixed points by flows of inhomogeneous random walks.