Fast N-Body Learning Nando de Freitas University of British Columbia.

Slides:



Advertisements
Similar presentations
Chapter 6 Differential Equations
Advertisements

Factorial Mixture of Gaussians and the Marginal Independence Model Ricardo Silva Joint work-in-progress with Zoubin Ghahramani.
7.4 – SOLVING SYSTEMS OF LINEAR EQUATIONS USING A SUBSTITUTION STRATEGY SYSTEMS OF LINEAR EQUATIONS.
Introduction to Belief Propagation and its Generalizations. Max Welling Donald Bren School of Information and Computer and Science University of California.
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.
Presenter: Yufan Liu November 17th,
Fast Computational Methods for Visually Guided Robots Maryam Mahdaviani, Nando de Freitas, Bob Fraser and Firas Hamze Department of Computer Science, University.
Spectral Clustering 指導教授 : 王聖智 S. J. Wang 學生 : 羅介暐 Jie-Wei Luo.
CS 584. Review n Systems of equations and finite element methods are related.
MA5233: Computational Mathematics
ECIV 301 Programming & Graphics Numerical Methods for Engineers Lecture 19 Solution of Linear System of Equations - Iterative Methods.
Semi-Supervised Learning in Gigantic Image Collections Rob Fergus (NYU) Yair Weiss (Hebrew U.) Antonio Torralba (MIT) TexPoint fonts used in EMF. Read.
High Performance Computing 1 Parallelization Strategies and Load Balancing Some material borrowed from lectures of J. Demmel, UC Berkeley.
Three Algorithms for Nonlinear Dimensionality Reduction Haixuan Yang Group Meeting Jan. 011, 2005.
MULTISCALE COMPUTATIONAL METHODS Achi Brandt The Weizmann Institute of Science UCLA
LESSON 3 – SOLVING SYSTEMS OF LINEAR EQUATIONS USING A SUBSTITUTION STRATEGY SYSTEMS OF LINEAR EQUATIONS.
Monday, March 23 Today's Objectives
CHAPTER 7-1 SOLVING SYSTEM OF EQUATIONS. WARM UP  Graph the following linear functions:  Y = 2x + 2  Y = 1/2x – 3  Y = -x - 1.
Distance Approximating Trees in Graphs
Math 3120 Differential Equations with Boundary Value Problems Chapter 2: First-Order Differential Equations Section 2-6: A Numerical Method.
Modelling and channel borrowing in mobile communications networks Richard J. Boucherie University of Twente Faculty of Mathematical Sciences University.
Linear Image Reconstruction Bart Janssen 13-11, 2007 Eindhoven.
Overview Particle filtering is a sequential Monte Carlo methodology in which the relevant probability distributions are iteratively estimated using the.
Physics Fluctuomatics / Applied Stochastic Process (Tohoku University) 1 Physical Fluctuomatics Applied Stochastic Process 9th Belief propagation Kazuyuki.
Math /4.2/4.3 – Solving Systems of Linear Equations 1.
Randomized Algorithms for Bayesian Hierarchical Clustering
-Arnaud Doucet, Nando de Freitas et al, UAI
Time-dependent Schrodinger Equation Numerical solution of the time-independent equation is straightforward constant energy solutions do not require us.
1 Dorit Aharonov Hebrew Univ. & UC Berkeley Adiabatic Quantum Computation.
Solving Linear Inequalities Lesson 5.5 linear inequality: _________________________________ ________________________________________________ solution of.
Quality of model and Error Analysis in Variational Data Assimilation François-Xavier LE DIMET Victor SHUTYAEV Université Joseph Fourier+INRIA Projet IDOPT,
Elimination Method: Solve the linear system. -8x + 3y=12 8x - 9y=12.
Suppose we are given a differential equation and initial condition: Then we can approximate the solution to the differential equation by its linearization.
1 6.1 Slope Fields and Euler's Method Objective: Solve differential equations graphically and numerically.
1 Beginning & Intermediate Algebra – Math 103 Math, Statistics & Physics.
to one side of an equation, When you do something.
+ Unit 1 – First degree equations and inequalities Chapter 3 – Systems of Equation and Inequalities 3.1 – Solving Systems by Graphing.
Systems of Equations and Inequalities Advanced Math Chapter 9.
Value Function Approximation with Diffusion Wavelets and Laplacian Eigenfunctions by S. Mahadevan & M. Maggioni Discussion led by Qi An ECE, Duke University.
CS Statistical Machine learning Lecture 25 Yuan (Alan) Qi Purdue CS Nov
Daphne Koller Overview Conditional Probability Queries Probabilistic Graphical Models Inference.
3.1 Graphing Systems of Equations Objective – To be able to solve and graph systems of linear equations. State Standard – 2.0 Students solve systems of.
Numerical Solutions of Partial Differential Equations CHAPTER 16.
Notes Over 3.1 Solving a System Graphically Graph the linear system and estimate the solution. Then check the solution algebraically.
EXAMPLE 1 Solve a system graphically Graph the linear system and estimate the solution. Then check the solution algebraically. 4x + y = 8 2x – 3y = 18.
10 October, 2007 University of Glasgow 1 EM Algorithm with Markov Chain Monte Carlo Method for Bayesian Image Analysis Kazuyuki Tanaka Graduate School.
Spectral Methods for Dimensionality
Differential Equations
3.3 – Solving Systems of Inequalities by Graphing
Systems of Nonlinear Equations
Markov Networks.
Ellipse Fitting COMP 4900C Winter 2008.
Linear Systems.
Solve a system of linear equation in two variables
Warm Up #3 1. Evaluate 5x + 2y for x = 2 and y = –4. 2 ANSWER
Singular Value Decomposition SVD
Graduate School of Information Sciences, Tohoku University
Differential Equations
Expectation-Maximization & Belief Propagation
Dynamical mean field theory: In practice
5.1 Solving Systems of Equations by Graphing
Solving Linear Equations by Graphing
 = N  N matrix multiplication N = 3 matrix N = 3 matrix N = 3 matrix
Systems of Equations Solve by Graphing.
Solve the linear system.
Markov Networks.
Linear and Nonlinear Systems of Equations
Linear and Nonlinear Systems of Equations
Introduction CSE 541.
Kazuyuki Tanaka Graduate School of Information Sciences
Presentation transcript:

Fast N-Body Learning Nando de Freitas University of British Columbia

Historical Perspective Non-iterative or “direct” methods for eigenvalue problems and linear systems of equations require O(N 3 ) operations. Let's look at the history of what has been regarded as large N: 1950: N= : N= : N= : N=20000 So over the course of 45 years N has increased by a factor of However, the speed of computers has increased by a factor of From this the O(N 3 ) bottleneck is evident. If only we could reduce the cost to O(N) – sigh!

Krylov for Eigen-Problems

Krylov for Systems of Equations

N-Body Problems in Learning Sum-kernel problem: Max-kernel problem:

N-Body Problems

Obvious applications of N-body Learning Exact and approximate message propagation. Markov chain Monte Carlo Gaussian processes, Wishart processes and Laplace processes. Spectral learning: eigenmaps, SNE, NCUTS, ranking on manifolds, … (even if using Nystrom) Reinforcement learning. The E step.

Kernel-(fill in your favourite name). Rao-Blackwellised Monte Carlo. Nearest neighbour methods. Some types of boosting. Computer graphics. EM, fluid dynamics, gravitation, quantum systems. … and much more ! Obvious applications of N-body Learning

Illustrative Example: Zhu, Lafferty & Zoubin

Illustrative Example Energy function using the graph Laplacian: Easy, but … a big linear system: Naïve iterative solution:

Illustrative Example We have solved a Gaussian process (where the covariance is the inverse graph Laplacian in O(N).

Illustrative Example

Message propagation Whether it’s exact: … or approximate:

Fast Methods in this Workshop

Fast Multipole Methods

Recursive Tree Structures

Distance Transform m(j) = min ( w(i) + d(i,j) ) i

In this workshop You’ll encounter tutorials on fast methods from the people who’ve been developing them. You’re likely to see encounter people arguing over error bounds, implementation strategies, applications and many more things. You’ll see statistics, learning, data structures and numerical computation come together. You’ll dream of the powder up on the hill.