Monte Carlo for Spatial Models

Size: px
Start display at page:

Download "Monte Carlo for Spatial Models"

Transcription

1 Monte Carlo for Spatial Models Murali Haran Department of Statistics Penn State University Penn State Computational Science Lectures April 2007

2 Spatial Models Lots of scientific questions involve analyzing data that are spatially dependent: Data points close together are more closely related (dependent) than data further away. Examples: Concentrations of PM2.5 (air pollutants) across the U.S. Disease rates by county. Abundance of plant/animal species across Pennsylvania. Space does not always mean physical distance. Similar ideas are applicable to many other research areas: Machine learning: two objects may be close in feature space. Approximations to computationally expensive computer models.

3 Geostatistical Data: Examples Wheat flowering dates in North Dakota (below): Courtesy Plant Pathology, PSU and North Dakota State. Other e.g. Concentrations of PM2.5 (pollutants) across the U.S.

4 Areal and Lattice Data: Examples Minnesota breast cancer by county: observed expected counts Courtesy MN Cancer Surveillance System, Dept. of Health Other e.g.pixel values from remote sensing e.g. PA forest cover.

5 Spatial Models: General Ideas We will focus on geostatistical data (process observed at points). Want a spatial process over a region. Simplest way to define a joint distribution is to assign a joint normal distribution to a finite set of points in that region. The set of points=location of observations and locations where we want to predict. Note: want this to be true (want a joint normal distribution) for any finite set of points in this region. This is therefore an infinite dimensional distribution called a Gaussian process.

6 Spatial Models: General Ideas (Contd.) Consider a joint distribution for 3 locations: Multivariate Normal: z 1 µ σ 11 σ 12 σ 13 z 2 N µ, σ 21 σ 22 σ 23. µ σ 31 σ 32 σ 33 z 3 We want the dependence (characterized by the covariance matrix) to be related to the distance between the locations. One possibility: Σ ij = ψ exp( ( s i s j )/φ) where s i is the location of the ith observation. So if the distance between ith and jth locations is large, Σ ij will be small. For example: if ψ = 2, φ = 0.4, location 1= (0,0), location 2=(1,2) then σ 12 = 2 exp( (5/0.4)).

7 Inference for Spatial Models A linear Gaussian process model with exponential covariance: Let Z (s) be spatial process s and X(s) be some covariate at location. For example: Z (s)= pollutant concentration at location s, X(s)= elevation at location s. Simple model: Z (s) = βx(s) + ɛ(s), where β is the regression coefficient and ɛ(s) is the error. If ɛ = (ɛ(s 1 ),..., ɛ(s N )) T, then ɛ N(0, Σ) where Σ ij = ψ exp( ( s i s j )/φ). We have a statistical model connecting data (Z, X) to parameters Θ = (β, ψ, φ). Inferred (estimated) Θ can be used to predict Z (s ) at any location s in region.

8 Inference for Spatial Models (contd.) Likelihood, L(Z; Θ) connects observations (Z) to parameters (Θ). It is proportional to multivariate normal density N(βX, Σ) for our example. Frequentist approach: maximize L(Z; Θ) w.r.t. Θ to obtain ˆΘ, the maximum likelihood estimate. Bayesian approach: treat Θ as random variable(s) and specify prior distribution f (Θ). Inference is based on posterior distribution, π(θ Z) = L(Z Θ)f (Θ) L(Z Θ)f (Θ) L(Z Θ)f (Θ)dΘ (This is just an application of Bayes rule)

9 Bayesian Inference for Spatial Models Z (s ) (estimate at a new location) is inferred from the posterior predictive distribution. π(z (s ) Z) = π(z (s ), Θ Z)dΘ = π(z (s ) Θ, Z)π(Θ Z)dΘ which is clearly obtained via the posterior distribution of the parameters Θ (given the observations Z). Note that this approach automatically propogates the variability associated with our inference about the parameters Θ. If we are less sure about Θ, that uncertainty is reflected in our estimate of Z (s ).

10 Inference for Spatial Models: Example Figure: North Dakota flowering dates, 2005: red: <July 6, orange: July 6 to July 12, brown:july 13 to July 18, yellow: after July 18. Left: Flowering survey data. Right: Posterior mean, E π (Z (s ) Z), based on the distribution π(z (s ) Z) at unobserved locations s.

11 Monte Carlo for Inference All inference for the model is based on the posterior (π). For e.g. E π (φ Z), the posterior expectation of parameter φ. In general we are interested in expectations of the form: E π g = g(x)π(x)dx Integral is too hard, so use sample based inference. We simulate X 1,..., X N from the distribution π. Use sample average: N i=1 g(x i)/n. In principle, if we have enough samples (large enough N), we can answer any question of interest. Example: What is the probability that φ > 0.8? Answer: Count the proportion of times sampled φ > 0.8.

12 Monte Carlo with iid samples Assume X 1,..., X N iid π. Strong Law holds: If E π g < then ḡ N = N g(x i )/N E π g as N. i=1 Central Limit Theorem: If E π g 2 < we have N(ḡN E π g) N(0, σ 2 ) Easy to estimate σ 2 using sample variance (ˆσ 2 ). Estimate accuracy of our estimate by ˆσ 2 /N. N is large enough when ˆσ 2 /N is small enough. Note that X i s can be multidimensional (accuracy is unaffected by the dimension of the problem here.)

13 Markov chain Monte Carlo Life is simple with i.i.d. Monte Carlo. Generally very difficult to draw i.i.d. samples from π, especially in high dimensions, complicated distributions. Not considered an option for spatial models in general. More general approach: Metropolis-Hastings algorithm. Start with an initial value X0. For i = 2 to N: Propose a value X for X i based on X i 1. Set X i = X with M-H probability depending on X i, X, the proposal distribution and the target distribution (π). The Markov chain X 1,..., X N has stationary distribution π (roughly: for large values of N, X N is approximately distributed according to π.)

14 Markov chain Monte Carlo (contd.) Use X 1,..., X N as before to estimate E π g. Strong Law holds if E π ( g ) <. ḡ N = N g(x i )/N E π g as N i=1 We also need technical conditions on the Markov chain but these typically hold by construction. It appears as if we have the same situation as in the i.i.d. case.

15 Markov chain Monte Carlo: Complications 1. The Central Limit Theorem may not hold. Hard to know when it does. 2. X i s are not i.i.d. so hard to estimate variance hard to rigorously assess the accuracy of our estimates. 3. We do not know how long to run our Markov chain. (How do we determine N for our sample X 1,..., X N?) 4. Not clear how to construct an efficient chain/algorithm for each new model,data set (Metropolis-Hastings only provides a general recipe.) Implication: Every time a user wants to fit a complex (e.g. spatial) model, needs to spend a lot of time tuning the algorithm. Also no guarantees about the accuracy of the estimates.

16 Some approaches I have considered 1. Exact/perfect sampling: avoid all MCMC issues by constructing samplers that produce i.i.d. draws. 2. Fast mixing algorithms using heavy-tailed approximations, transformations etc. Some samplers with known (good) theoretical properties. Others that appear to work well in practice (based on empirical studies). 3. Monte Carlo standard errors: Consistent and easy-to-use estimator for assessing standard errors. When standard errors are below a threshold, stop the MCMC sampler. Note: These approaches are not mutually exclusive.

17 Toy example: Normal density x Density x y Left (approximation-based): Green line is heavy-tailed approximation (t-density), red line is target (normal density.) Use heavy-tailed approximation as proposal. Right (transformation-based): Bounded 2D region corresponds to transformation of 1D normal (y/x normal density.) Simulate in transformed space.

18 For spatial models Linear spatial model where inference is based on π(z (s ) Z) = π(z (s ), Θ Z)dΘ allows us to deal with the simulation in stages: 1. Simulate Θ π(θ Z). 2. Simulate Z (s ) π(z (s ) Θ, Z), which is just a draw from a multivariate normal. With this approach: Even though inference may be for a high-dimensional distribution, only step 1 is problematic. Step 1 involves few dimensions (typically around 4 though it could have more if there are many predictors.) If we construct a very good heavy-tailed approximation to propose values for MCMC or simulate on an appropriate transformed space, we will have an algorithm with good properties.

19 Spatial generalized linear models What if data are non-gaussian (e.g. 0-1 or count data)? Hierarchical modeling is not a problem. For example: Stage 1: Z (s) µ(s) Poisson(E(s) exp(µ(s))). Stage 2: Now model µ(s) Θ as a Gaussian process. Stage 3: Priors for Θ. This specification destroys our ability to simulate easily: 1. Now need to simulate Θ, µ (s 1 ),..., µ (s n ) π(θ, µ (s 1 ),..., µ (s n ) Z). 2. Simulate Z (s ) π(z (s ) Θ, µ (s 1 ),..., µ (s n ), Z). Step 1 is now more complicated and dimensions of distribution # of observations. One approach: Approximate this model by a linear hierarchical model.use samples from approximation to obtain draws from above distribution.

20 Some lessons learned Exact sampling: ideal situation but very hard to construct. Algorithms end up being very specialized. Fast mixing algorithms: best case when the algorithm has good theoretical properties and works well in practice (one does not imply the other.) Putting it all together: Using a good estimate of standard errors for an algorithm with good theoretical properties: Efficient algorithm. Easy, accurate assessment of standard errors. A simple rule for stopping the algorithm based on desired accuracy.

21 Some lessons learned (contd.) If the sampler gets stuck in one area of the sample space (if π has well separated modes.) Even estimates of standard error can be misleading. Hence, first requirement: good sampler. Very important to use as much information as possible about π when constructing a sampler. Exploit the structure of the model/distribution. For e.g. spatial models have a lot of structure. Utilize matrix algorithms (e.g. sparse matrix inversions, choleskis etc.) whenever possible. Be aware of possible multimodalities (obvious in some scenarios, not so obvious in others.) Always attempt to quantify the accuracy of your estimates.

22 Some references Accurate standard errors and a stopping rule for MCMC: Flegal, J.M., Haran, M. and Jones, G.L. Markov chain Monte Carlo: Can We Trust the Third Decimal Place? R code for estimating errors easily via consistent batch means : mharan/batchmeans.r Experiments with block updating of parameters: Haran, M., Hodges, J.S., and Carlin, B.P. (2003), Accelerating computation in Markov random field models for spatial data via structured MCMC, J.Comp.Graph.Stat.. Exact sampling using a fast mixing MCMC algorithm: Haran, M. and Tierney, L., Perfect sampling for a Markov random field model.

MCMC Methods for data modeling

MCMC Methods for data modeling MCMC Methods for data modeling Kenneth Scerri Department of Automatic Control and Systems Engineering Introduction 1. Symposium on Data Modelling 2. Outline: a. Definition and uses of MCMC b. MCMC algorithms

More information

Bayesian Spatiotemporal Modeling with Hierarchical Spatial Priors for fmri

Bayesian Spatiotemporal Modeling with Hierarchical Spatial Priors for fmri Bayesian Spatiotemporal Modeling with Hierarchical Spatial Priors for fmri Galin L. Jones 1 School of Statistics University of Minnesota March 2015 1 Joint with Martin Bezener and John Hughes Experiment

More information

Parallel Multivariate Slice Sampling

Parallel Multivariate Slice Sampling Parallel Multivariate Slice Sampling Matthew M. Tibbits Department of Statistics Pennsylvania State University mmt143@stat.psu.edu Murali Haran Department of Statistics Pennsylvania State University mharan@stat.psu.edu

More information

MCMC Diagnostics. Yingbo Li MATH Clemson University. Yingbo Li (Clemson) MCMC Diagnostics MATH / 24

MCMC Diagnostics. Yingbo Li MATH Clemson University. Yingbo Li (Clemson) MCMC Diagnostics MATH / 24 MCMC Diagnostics Yingbo Li Clemson University MATH 9810 Yingbo Li (Clemson) MCMC Diagnostics MATH 9810 1 / 24 Convergence to Posterior Distribution Theory proves that if a Gibbs sampler iterates enough,

More information

Overview. Monte Carlo Methods. Statistics & Bayesian Inference Lecture 3. Situation At End Of Last Week

Overview. Monte Carlo Methods. Statistics & Bayesian Inference Lecture 3. Situation At End Of Last Week Statistics & Bayesian Inference Lecture 3 Joe Zuntz Overview Overview & Motivation Metropolis Hastings Monte Carlo Methods Importance sampling Direct sampling Gibbs sampling Monte-Carlo Markov Chains Emcee

More information

INLA: an introduction

INLA: an introduction INLA: an introduction Håvard Rue 1 Norwegian University of Science and Technology Trondheim, Norway May 2009 1 Joint work with S.Martino (Trondheim) and N.Chopin (Paris) Latent Gaussian models Background

More information

Chapter 1. Introduction

Chapter 1. Introduction Chapter 1 Introduction A Monte Carlo method is a compuational method that uses random numbers to compute (estimate) some quantity of interest. Very often the quantity we want to compute is the mean of

More information

Markov chain Monte Carlo methods

Markov chain Monte Carlo methods Markov chain Monte Carlo methods (supplementary material) see also the applet http://www.lbreyer.com/classic.html February 9 6 Independent Hastings Metropolis Sampler Outline Independent Hastings Metropolis

More information

Variability in Annual Temperature Profiles

Variability in Annual Temperature Profiles Variability in Annual Temperature Profiles A Multivariate Spatial Analysis of Regional Climate Model Output Tamara Greasby, Stephan Sain Institute for Mathematics Applied to Geosciences, National Center

More information

Bayesian Statistics Group 8th March Slice samplers. (A very brief introduction) The basic idea

Bayesian Statistics Group 8th March Slice samplers. (A very brief introduction) The basic idea Bayesian Statistics Group 8th March 2000 Slice samplers (A very brief introduction) The basic idea lacements To sample from a distribution, simply sample uniformly from the region under the density function

More information

Quantitative Biology II!

Quantitative Biology II! Quantitative Biology II! Lecture 3: Markov Chain Monte Carlo! March 9, 2015! 2! Plan for Today!! Introduction to Sampling!! Introduction to MCMC!! Metropolis Algorithm!! Metropolis-Hastings Algorithm!!

More information

1 Methods for Posterior Simulation

1 Methods for Posterior Simulation 1 Methods for Posterior Simulation Let p(θ y) be the posterior. simulation. Koop presents four methods for (posterior) 1. Monte Carlo integration: draw from p(θ y). 2. Gibbs sampler: sequentially drawing

More information

INLA: Integrated Nested Laplace Approximations

INLA: Integrated Nested Laplace Approximations INLA: Integrated Nested Laplace Approximations John Paige Statistics Department University of Washington October 10, 2017 1 The problem Markov Chain Monte Carlo (MCMC) takes too long in many settings.

More information

Missing Data Analysis for the Employee Dataset

Missing Data Analysis for the Employee Dataset Missing Data Analysis for the Employee Dataset 67% of the observations have missing values! Modeling Setup Random Variables: Y i =(Y i1,...,y ip ) 0 =(Y i,obs, Y i,miss ) 0 R i =(R i1,...,r ip ) 0 ( 1

More information

Monte Carlo Methods and Statistical Computing: My Personal E

Monte Carlo Methods and Statistical Computing: My Personal E Monte Carlo Methods and Statistical Computing: My Personal Experience Department of Mathematics & Statistics Indian Institute of Technology Kanpur November 29, 2014 Outline Preface 1 Preface 2 3 4 5 6

More information

The Pennsylvania State University The Graduate School PARALLEL MULTIVARIATE SLICE SAMPLING. A Thesis in Statistics by Matthew M.

The Pennsylvania State University The Graduate School PARALLEL MULTIVARIATE SLICE SAMPLING. A Thesis in Statistics by Matthew M. The Pennsylvania State University The Graduate School PARALLEL MULTIVARIATE SLICE SAMPLING A Thesis in Statistics by Matthew M. Tibbits c 2009 Matthew M. Tibbits Submitted in Partial Fulfillment of the

More information

A GENERAL GIBBS SAMPLING ALGORITHM FOR ANALYZING LINEAR MODELS USING THE SAS SYSTEM

A GENERAL GIBBS SAMPLING ALGORITHM FOR ANALYZING LINEAR MODELS USING THE SAS SYSTEM A GENERAL GIBBS SAMPLING ALGORITHM FOR ANALYZING LINEAR MODELS USING THE SAS SYSTEM Jayawant Mandrekar, Daniel J. Sargent, Paul J. Novotny, Jeff A. Sloan Mayo Clinic, Rochester, MN 55905 ABSTRACT A general

More information

Probabilistic Graphical Models

Probabilistic Graphical Models Probabilistic Graphical Models Lecture 17 EM CS/CNS/EE 155 Andreas Krause Announcements Project poster session on Thursday Dec 3, 4-6pm in Annenberg 2 nd floor atrium! Easels, poster boards and cookies

More information

Nested Sampling: Introduction and Implementation

Nested Sampling: Introduction and Implementation UNIVERSITY OF TEXAS AT SAN ANTONIO Nested Sampling: Introduction and Implementation Liang Jing May 2009 1 1 ABSTRACT Nested Sampling is a new technique to calculate the evidence, Z = P(D M) = p(d θ, M)p(θ

More information

Approximate Bayesian Computation. Alireza Shafaei - April 2016

Approximate Bayesian Computation. Alireza Shafaei - April 2016 Approximate Bayesian Computation Alireza Shafaei - April 2016 The Problem Given a dataset, we are interested in. The Problem Given a dataset, we are interested in. The Problem Given a dataset, we are interested

More information

Statistical Matching using Fractional Imputation

Statistical Matching using Fractional Imputation Statistical Matching using Fractional Imputation Jae-Kwang Kim 1 Iowa State University 1 Joint work with Emily Berg and Taesung Park 1 Introduction 2 Classical Approaches 3 Proposed method 4 Application:

More information

ADAPTIVE METROPOLIS-HASTINGS SAMPLING, OR MONTE CARLO KERNEL ESTIMATION

ADAPTIVE METROPOLIS-HASTINGS SAMPLING, OR MONTE CARLO KERNEL ESTIMATION ADAPTIVE METROPOLIS-HASTINGS SAMPLING, OR MONTE CARLO KERNEL ESTIMATION CHRISTOPHER A. SIMS Abstract. A new algorithm for sampling from an arbitrary pdf. 1. Introduction Consider the standard problem of

More information

Analysis of Incomplete Multivariate Data

Analysis of Incomplete Multivariate Data Analysis of Incomplete Multivariate Data J. L. Schafer Department of Statistics The Pennsylvania State University USA CHAPMAN & HALL/CRC A CR.C Press Company Boca Raton London New York Washington, D.C.

More information

A Nonparametric Bayesian Approach to Detecting Spatial Activation Patterns in fmri Data

A Nonparametric Bayesian Approach to Detecting Spatial Activation Patterns in fmri Data A Nonparametric Bayesian Approach to Detecting Spatial Activation Patterns in fmri Data Seyoung Kim, Padhraic Smyth, and Hal Stern Bren School of Information and Computer Sciences University of California,

More information

Mixture Models and the EM Algorithm

Mixture Models and the EM Algorithm Mixture Models and the EM Algorithm Padhraic Smyth, Department of Computer Science University of California, Irvine c 2017 1 Finite Mixture Models Say we have a data set D = {x 1,..., x N } where x i is

More information

An Introduction to Markov Chain Monte Carlo

An Introduction to Markov Chain Monte Carlo An Introduction to Markov Chain Monte Carlo Markov Chain Monte Carlo (MCMC) refers to a suite of processes for simulating a posterior distribution based on a random (ie. monte carlo) process. In other

More information

Hierarchical Modelling for Large Spatial Datasets

Hierarchical Modelling for Large Spatial Datasets Hierarchical Modelling for Large Spatial Datasets Sudipto Banerjee 1 and Andrew O. Finley 2 1 Department of Forestry & Department of Geography, Michigan State University, Lansing Michigan, U.S.A. 2 Biostatistics,

More information

Lecture 27, April 24, Reading: See class website. Nonparametric regression and kernel smoothing. Structured sparse additive models (GroupSpAM)

Lecture 27, April 24, Reading: See class website. Nonparametric regression and kernel smoothing. Structured sparse additive models (GroupSpAM) School of Computer Science Probabilistic Graphical Models Structured Sparse Additive Models Junming Yin and Eric Xing Lecture 7, April 4, 013 Reading: See class website 1 Outline Nonparametric regression

More information

Global modelling of air pollution using multiple data sources

Global modelling of air pollution using multiple data sources Global modelling of air pollution using multiple data sources Matthew Thomas SAMBa, University of Bath Email: M.L.Thomas@bath.ac.uk November 11, 015 1/ 3 OUTLINE Motivation Data Sources Existing Approaches

More information

STAT 725 Notes Monte Carlo Integration

STAT 725 Notes Monte Carlo Integration STAT 725 Notes Monte Carlo Integration Two major classes of numerical problems arise in statistical inference: optimization and integration. We have already spent some time discussing different optimization

More information

08 An Introduction to Dense Continuous Robotic Mapping

08 An Introduction to Dense Continuous Robotic Mapping NAVARCH/EECS 568, ROB 530 - Winter 2018 08 An Introduction to Dense Continuous Robotic Mapping Maani Ghaffari March 14, 2018 Previously: Occupancy Grid Maps Pose SLAM graph and its associated dense occupancy

More information

From Bayesian Analysis of Item Response Theory Models Using SAS. Full book available for purchase here.

From Bayesian Analysis of Item Response Theory Models Using SAS. Full book available for purchase here. From Bayesian Analysis of Item Response Theory Models Using SAS. Full book available for purchase here. Contents About this Book...ix About the Authors... xiii Acknowledgments... xv Chapter 1: Item Response

More information

Bayesian Estimation for Skew Normal Distributions Using Data Augmentation

Bayesian Estimation for Skew Normal Distributions Using Data Augmentation The Korean Communications in Statistics Vol. 12 No. 2, 2005 pp. 323-333 Bayesian Estimation for Skew Normal Distributions Using Data Augmentation Hea-Jung Kim 1) Abstract In this paper, we develop a MCMC

More information

Variational Methods for Graphical Models

Variational Methods for Graphical Models Chapter 2 Variational Methods for Graphical Models 2.1 Introduction The problem of probabb1istic inference in graphical models is the problem of computing a conditional probability distribution over the

More information

Theoretical Concepts of Machine Learning

Theoretical Concepts of Machine Learning Theoretical Concepts of Machine Learning Part 2 Institute of Bioinformatics Johannes Kepler University, Linz, Austria Outline 1 Introduction 2 Generalization Error 3 Maximum Likelihood 4 Noise Models 5

More information

Expectation-Maximization Methods in Population Analysis. Robert J. Bauer, Ph.D. ICON plc.

Expectation-Maximization Methods in Population Analysis. Robert J. Bauer, Ph.D. ICON plc. Expectation-Maximization Methods in Population Analysis Robert J. Bauer, Ph.D. ICON plc. 1 Objective The objective of this tutorial is to briefly describe the statistical basis of Expectation-Maximization

More information

Computer Vision Group Prof. Daniel Cremers. 4. Probabilistic Graphical Models Directed Models

Computer Vision Group Prof. Daniel Cremers. 4. Probabilistic Graphical Models Directed Models Prof. Daniel Cremers 4. Probabilistic Graphical Models Directed Models The Bayes Filter (Rep.) (Bayes) (Markov) (Tot. prob.) (Markov) (Markov) 2 Graphical Representation (Rep.) We can describe the overall

More information

Stochastic Simulation: Algorithms and Analysis

Stochastic Simulation: Algorithms and Analysis Soren Asmussen Peter W. Glynn Stochastic Simulation: Algorithms and Analysis et Springer Contents Preface Notation v xii I What This Book Is About 1 1 An Illustrative Example: The Single-Server Queue 1

More information

FMA901F: Machine Learning Lecture 3: Linear Models for Regression. Cristian Sminchisescu

FMA901F: Machine Learning Lecture 3: Linear Models for Regression. Cristian Sminchisescu FMA901F: Machine Learning Lecture 3: Linear Models for Regression Cristian Sminchisescu Machine Learning: Frequentist vs. Bayesian In the frequentist setting, we seek a fixed parameter (vector), with value(s)

More information

Calibration and emulation of TIE-GCM

Calibration and emulation of TIE-GCM Calibration and emulation of TIE-GCM Serge Guillas School of Mathematics Georgia Institute of Technology Jonathan Rougier University of Bristol Big Thanks to Crystal Linkletter (SFU-SAMSI summer school)

More information

Computer vision: models, learning and inference. Chapter 10 Graphical Models

Computer vision: models, learning and inference. Chapter 10 Graphical Models Computer vision: models, learning and inference Chapter 10 Graphical Models Independence Two variables x 1 and x 2 are independent if their joint probability distribution factorizes as Pr(x 1, x 2 )=Pr(x

More information

A noninformative Bayesian approach to small area estimation

A noninformative Bayesian approach to small area estimation A noninformative Bayesian approach to small area estimation Glen Meeden School of Statistics University of Minnesota Minneapolis, MN 55455 glen@stat.umn.edu September 2001 Revised May 2002 Research supported

More information

CS839: Probabilistic Graphical Models. Lecture 10: Learning with Partially Observed Data. Theo Rekatsinas

CS839: Probabilistic Graphical Models. Lecture 10: Learning with Partially Observed Data. Theo Rekatsinas CS839: Probabilistic Graphical Models Lecture 10: Learning with Partially Observed Data Theo Rekatsinas 1 Partially Observed GMs Speech recognition 2 Partially Observed GMs Evolution 3 Partially Observed

More information

Clustering Relational Data using the Infinite Relational Model

Clustering Relational Data using the Infinite Relational Model Clustering Relational Data using the Infinite Relational Model Ana Daglis Supervised by: Matthew Ludkin September 4, 2015 Ana Daglis Clustering Data using the Infinite Relational Model September 4, 2015

More information

Issues in MCMC use for Bayesian model fitting. Practical Considerations for WinBUGS Users

Issues in MCMC use for Bayesian model fitting. Practical Considerations for WinBUGS Users Practical Considerations for WinBUGS Users Kate Cowles, Ph.D. Department of Statistics and Actuarial Science University of Iowa 22S:138 Lecture 12 Oct. 3, 2003 Issues in MCMC use for Bayesian model fitting

More information

Image analysis. Computer Vision and Classification Image Segmentation. 7 Image analysis

Image analysis. Computer Vision and Classification Image Segmentation. 7 Image analysis 7 Computer Vision and Classification 413 / 458 Computer Vision and Classification The k-nearest-neighbor method The k-nearest-neighbor (knn) procedure has been used in data analysis and machine learning

More information

Clustering. Mihaela van der Schaar. January 27, Department of Engineering Science University of Oxford

Clustering. Mihaela van der Schaar. January 27, Department of Engineering Science University of Oxford Department of Engineering Science University of Oxford January 27, 2017 Many datasets consist of multiple heterogeneous subsets. Cluster analysis: Given an unlabelled data, want algorithms that automatically

More information

Bayesian Inference for Sample Surveys

Bayesian Inference for Sample Surveys Bayesian Inference for Sample Surveys Trivellore Raghunathan (Raghu) Director, Survey Research Center Professor of Biostatistics University of Michigan Distinctive features of survey inference 1. Primary

More information

Convexization in Markov Chain Monte Carlo

Convexization in Markov Chain Monte Carlo in Markov Chain Monte Carlo 1 IBM T. J. Watson Yorktown Heights, NY 2 Department of Aerospace Engineering Technion, Israel August 23, 2011 Problem Statement MCMC processes in general are governed by non

More information

Non-parametric Approximate Bayesian Computation for Expensive Simulators

Non-parametric Approximate Bayesian Computation for Expensive Simulators Non-parametric Approximate Bayesian Computation for Expensive Simulators Steven Laan 6036031 Master s thesis 42 EC Master s programme Artificial Intelligence University of Amsterdam Supervisor Ted Meeds

More information

Statistical techniques for data analysis in Cosmology

Statistical techniques for data analysis in Cosmology Statistical techniques for data analysis in Cosmology arxiv:0712.3028; arxiv:0911.3105 Numerical recipes (the bible ) Licia Verde ICREA & ICC UB-IEEC http://icc.ub.edu/~liciaverde outline Lecture 1: Introduction

More information

Exam Issued: May 29, 2017, 13:00 Hand in: May 29, 2017, 16:00

Exam Issued: May 29, 2017, 13:00 Hand in: May 29, 2017, 16:00 P. Hadjidoukas, C. Papadimitriou ETH Zentrum, CTL E 13 CH-8092 Zürich High Performance Computing for Science and Engineering II Exam Issued: May 29, 2017, 13:00 Hand in: May 29, 2017, 16:00 Spring semester

More information

Global modelling of air pollution using multiple data sources

Global modelling of air pollution using multiple data sources Global modelling of air pollution using multiple data sources Matthew Thomas M.L.Thomas@bath.ac.uk Supervised by Dr. Gavin Shaddick In collaboration with IHME and WHO June 14, 2016 1/ 1 MOTIVATION Air

More information

Computer Vision Group Prof. Daniel Cremers. 4. Probabilistic Graphical Models Directed Models

Computer Vision Group Prof. Daniel Cremers. 4. Probabilistic Graphical Models Directed Models Prof. Daniel Cremers 4. Probabilistic Graphical Models Directed Models The Bayes Filter (Rep.) (Bayes) (Markov) (Tot. prob.) (Markov) (Markov) 2 Graphical Representation (Rep.) We can describe the overall

More information

Sampling informative/complex a priori probability distributions using Gibbs sampling assisted by sequential simulation

Sampling informative/complex a priori probability distributions using Gibbs sampling assisted by sequential simulation Sampling informative/complex a priori probability distributions using Gibbs sampling assisted by sequential simulation Thomas Mejer Hansen, Klaus Mosegaard, and Knud Skou Cordua 1 1 Center for Energy Resources

More information

Learning from Data: Adaptive Basis Functions

Learning from Data: Adaptive Basis Functions Learning from Data: Adaptive Basis Functions November 21, 2005 http://www.anc.ed.ac.uk/ amos/lfd/ Neural Networks Hidden to output layer - a linear parameter model But adapt the features of the model.

More information

CS281 Section 9: Graph Models and Practical MCMC

CS281 Section 9: Graph Models and Practical MCMC CS281 Section 9: Graph Models and Practical MCMC Scott Linderman November 11, 213 Now that we have a few MCMC inference algorithms in our toolbox, let s try them out on some random graph models. Graphs

More information

Integrating auxiliary data in optimal spatial design for species distribution mapping

Integrating auxiliary data in optimal spatial design for species distribution mapping Integrating auxiliary data in optimal spatial design for species distribution mapping Brian Reich, Krishna Pacifici and Jon Stallings North Carolina State University Reich + Pacifici + Stallings Optimal

More information

Recap: The E-M algorithm. Biostatistics 615/815 Lecture 22: Gibbs Sampling. Recap - Local minimization methods

Recap: The E-M algorithm. Biostatistics 615/815 Lecture 22: Gibbs Sampling. Recap - Local minimization methods Recap: The E-M algorithm Biostatistics 615/815 Lecture 22: Gibbs Sampling Expectation step (E-step) Given the current estimates of parameters λ (t), calculate the conditional distribution of latent variable

More information

Modeling Criminal Careers as Departures From a Unimodal Population Age-Crime Curve: The Case of Marijuana Use

Modeling Criminal Careers as Departures From a Unimodal Population Age-Crime Curve: The Case of Marijuana Use Modeling Criminal Careers as Departures From a Unimodal Population Curve: The Case of Marijuana Use Donatello Telesca, Elena A. Erosheva, Derek A. Kreader, & Ross Matsueda April 15, 2014 extends Telesca

More information

Machine Learning. Sourangshu Bhattacharya

Machine Learning. Sourangshu Bhattacharya Machine Learning Sourangshu Bhattacharya Bayesian Networks Directed Acyclic Graph (DAG) Bayesian Networks General Factorization Curve Fitting Re-visited Maximum Likelihood Determine by minimizing sum-of-squares

More information

10703 Deep Reinforcement Learning and Control

10703 Deep Reinforcement Learning and Control 10703 Deep Reinforcement Learning and Control Russ Salakhutdinov Machine Learning Department rsalakhu@cs.cmu.edu Policy Gradient I Used Materials Disclaimer: Much of the material and slides for this lecture

More information

Probabilistic Graphical Models

Probabilistic Graphical Models Overview of Part Two Probabilistic Graphical Models Part Two: Inference and Learning Christopher M. Bishop Exact inference and the junction tree MCMC Variational methods and EM Example General variational

More information

Level-set MCMC Curve Sampling and Geometric Conditional Simulation

Level-set MCMC Curve Sampling and Geometric Conditional Simulation Level-set MCMC Curve Sampling and Geometric Conditional Simulation Ayres Fan John W. Fisher III Alan S. Willsky February 16, 2007 Outline 1. Overview 2. Curve evolution 3. Markov chain Monte Carlo 4. Curve

More information

Graphical Models, Bayesian Method, Sampling, and Variational Inference

Graphical Models, Bayesian Method, Sampling, and Variational Inference Graphical Models, Bayesian Method, Sampling, and Variational Inference With Application in Function MRI Analysis and Other Imaging Problems Wei Liu Scientific Computing and Imaging Institute University

More information

Markov Chain Monte Carlo (part 1)

Markov Chain Monte Carlo (part 1) Markov Chain Monte Carlo (part 1) Edps 590BAY Carolyn J. Anderson Department of Educational Psychology c Board of Trustees, University of Illinois Spring 2018 Depending on the book that you select for

More information

BART STAT8810, Fall 2017

BART STAT8810, Fall 2017 BART STAT8810, Fall 2017 M.T. Pratola November 1, 2017 Today BART: Bayesian Additive Regression Trees BART: Bayesian Additive Regression Trees Additive model generalizes the single-tree regression model:

More information

Math 494: Mathematical Statistics

Math 494: Mathematical Statistics Math 494: Mathematical Statistics Instructor: Jimin Ding jmding@wustl.edu Department of Mathematics Washington University in St. Louis Class materials are available on course website (www.math.wustl.edu/

More information

A Survey of Statistical Models to Infer Consensus 3D Chromosomal Structure from Hi-C data

A Survey of Statistical Models to Infer Consensus 3D Chromosomal Structure from Hi-C data A Survey of Statistical Models to Infer Consensus 3D Chromosomal Structure from Hi-C data MEDHA UPPALA, University of California Los Angeles The spatial organization of the genomic material leads to interactions

More information

10.4 Linear interpolation method Newton s method

10.4 Linear interpolation method Newton s method 10.4 Linear interpolation method The next best thing one can do is the linear interpolation method, also known as the double false position method. This method works similarly to the bisection method by

More information

Efficient Feature Learning Using Perturb-and-MAP

Efficient Feature Learning Using Perturb-and-MAP Efficient Feature Learning Using Perturb-and-MAP Ke Li, Kevin Swersky, Richard Zemel Dept. of Computer Science, University of Toronto {keli,kswersky,zemel}@cs.toronto.edu Abstract Perturb-and-MAP [1] is

More information

Convergence and Efficiency of Adaptive MCMC. Jinyoung Yang

Convergence and Efficiency of Adaptive MCMC. Jinyoung Yang Convergence and Efficiency of Adaptive MCMC by Jinyoung Yang A thesis submitted in conformity with the requirements for the degree of Doctor of Philosophy Graduate Department of Statistical Sciences University

More information

Adaptive spatial resampling as a Markov chain Monte Carlo method for uncertainty quantification in seismic reservoir characterization

Adaptive spatial resampling as a Markov chain Monte Carlo method for uncertainty quantification in seismic reservoir characterization 1 Adaptive spatial resampling as a Markov chain Monte Carlo method for uncertainty quantification in seismic reservoir characterization Cheolkyun Jeong, Tapan Mukerji, and Gregoire Mariethoz Department

More information

D025 Geostatistical Stochastic Elastic Iinversion - An Efficient Method for Integrating Seismic and Well Data Constraints

D025 Geostatistical Stochastic Elastic Iinversion - An Efficient Method for Integrating Seismic and Well Data Constraints D025 Geostatistical Stochastic Elastic Iinversion - An Efficient Method for Integrating Seismic and Well Data Constraints P.R. Williamson (Total E&P USA Inc.), A.J. Cherrett* (Total SA) & R. Bornard (CGGVeritas)

More information

AMCMC: An R interface for adaptive MCMC

AMCMC: An R interface for adaptive MCMC AMCMC: An R interface for adaptive MCMC by Jeffrey S. Rosenthal * (February 2007) Abstract. We describe AMCMC, a software package for running adaptive MCMC algorithms on user-supplied density functions.

More information

10-701/15-781, Fall 2006, Final

10-701/15-781, Fall 2006, Final -7/-78, Fall 6, Final Dec, :pm-8:pm There are 9 questions in this exam ( pages including this cover sheet). If you need more room to work out your answer to a question, use the back of the page and clearly

More information

Rolling Markov Chain Monte Carlo

Rolling Markov Chain Monte Carlo Rolling Markov Chain Monte Carlo Din-Houn Lau Imperial College London Joint work with Axel Gandy 4 th July 2013 Predict final ranks of the each team. Updates quick update of predictions. Accuracy control

More information

CSCI 599 Class Presenta/on. Zach Levine. Markov Chain Monte Carlo (MCMC) HMM Parameter Es/mates

CSCI 599 Class Presenta/on. Zach Levine. Markov Chain Monte Carlo (MCMC) HMM Parameter Es/mates CSCI 599 Class Presenta/on Zach Levine Markov Chain Monte Carlo (MCMC) HMM Parameter Es/mates April 26 th, 2012 Topics Covered in this Presenta2on A (Brief) Review of HMMs HMM Parameter Learning Expecta2on-

More information

Performance of Sequential Imputation Method in Multilevel Applications

Performance of Sequential Imputation Method in Multilevel Applications Section on Survey Research Methods JSM 9 Performance of Sequential Imputation Method in Multilevel Applications Enxu Zhao, Recai M. Yucel New York State Department of Health, 8 N. Pearl St., Albany, NY

More information

Missing Data Analysis for the Employee Dataset

Missing Data Analysis for the Employee Dataset Missing Data Analysis for the Employee Dataset 67% of the observations have missing values! Modeling Setup For our analysis goals we would like to do: Y X N (X, 2 I) and then interpret the coefficients

More information

MSA101/MVE Lecture 5

MSA101/MVE Lecture 5 MSA101/MVE187 2017 Lecture 5 Petter Mostad Chalmers University September 12, 2017 1 / 15 Importance sampling MC integration computes h(x)f (x) dx where f (x) is a probability density function, by simulating

More information

RJaCGH, a package for analysis of

RJaCGH, a package for analysis of RJaCGH, a package for analysis of CGH arrays with Reversible Jump MCMC 1. CGH Arrays: Biological problem: Changes in number of DNA copies are associated to cancer activity. Microarray technology: Oscar

More information

GAMs semi-parametric GLMs. Simon Wood Mathematical Sciences, University of Bath, U.K.

GAMs semi-parametric GLMs. Simon Wood Mathematical Sciences, University of Bath, U.K. GAMs semi-parametric GLMs Simon Wood Mathematical Sciences, University of Bath, U.K. Generalized linear models, GLM 1. A GLM models a univariate response, y i as g{e(y i )} = X i β where y i Exponential

More information

Predictor Selection Algorithm for Bayesian Lasso

Predictor Selection Algorithm for Bayesian Lasso Predictor Selection Algorithm for Baesian Lasso Quan Zhang Ma 16, 2014 1 Introduction The Lasso [1] is a method in regression model for coefficients shrinkage and model selection. It is often used in the

More information

Graphical Models & HMMs

Graphical Models & HMMs Graphical Models & HMMs Henrik I. Christensen Robotics & Intelligent Machines @ GT Georgia Institute of Technology, Atlanta, GA 30332-0280 hic@cc.gatech.edu Henrik I. Christensen (RIM@GT) Graphical Models

More information

Linear Modeling with Bayesian Statistics

Linear Modeling with Bayesian Statistics Linear Modeling with Bayesian Statistics Bayesian Approach I I I I I Estimate probability of a parameter State degree of believe in specific parameter values Evaluate probability of hypothesis given the

More information

PSU Student Research Symposium 2017 Bayesian Optimization for Refining Object Proposals, with an Application to Pedestrian Detection Anthony D.

PSU Student Research Symposium 2017 Bayesian Optimization for Refining Object Proposals, with an Application to Pedestrian Detection Anthony D. PSU Student Research Symposium 2017 Bayesian Optimization for Refining Object Proposals, with an Application to Pedestrian Detection Anthony D. Rhodes 5/10/17 What is Machine Learning? Machine learning

More information

GAMES Webinar: Rendering Tutorial 2. Monte Carlo Methods. Shuang Zhao

GAMES Webinar: Rendering Tutorial 2. Monte Carlo Methods. Shuang Zhao GAMES Webinar: Rendering Tutorial 2 Monte Carlo Methods Shuang Zhao Assistant Professor Computer Science Department University of California, Irvine GAMES Webinar Shuang Zhao 1 Outline 1. Monte Carlo integration

More information

Homework. Gaussian, Bishop 2.3 Non-parametric, Bishop 2.5 Linear regression Pod-cast lecture on-line. Next lectures:

Homework. Gaussian, Bishop 2.3 Non-parametric, Bishop 2.5 Linear regression Pod-cast lecture on-line. Next lectures: Homework Gaussian, Bishop 2.3 Non-parametric, Bishop 2.5 Linear regression 3.0-3.2 Pod-cast lecture on-line Next lectures: I posted a rough plan. It is flexible though so please come with suggestions Bayes

More information

Bayesian Modelling with JAGS and R

Bayesian Modelling with JAGS and R Bayesian Modelling with JAGS and R Martyn Plummer International Agency for Research on Cancer Rencontres R, 3 July 2012 CRAN Task View Bayesian Inference The CRAN Task View Bayesian Inference is maintained

More information

Spatial Point Patterns

Spatial Point Patterns What is a point pattern? Spatial Point Patterns Sudipto Banerjee 1, Bradley P. Carlin 1 and Alan E. Gelfand 2 1 Division of Biostatistics, School of Public Health, University of Minnesota, Minneapolis,

More information

Modified Metropolis-Hastings algorithm with delayed rejection

Modified Metropolis-Hastings algorithm with delayed rejection Modified Metropolis-Hastings algorithm with delayed reection K.M. Zuev & L.S. Katafygiotis Department of Civil Engineering, Hong Kong University of Science and Technology, Hong Kong, China ABSTRACT: The

More information

Discussion on Bayesian Model Selection and Parameter Estimation in Extragalactic Astronomy by Martin Weinberg

Discussion on Bayesian Model Selection and Parameter Estimation in Extragalactic Astronomy by Martin Weinberg Discussion on Bayesian Model Selection and Parameter Estimation in Extragalactic Astronomy by Martin Weinberg Phil Gregory Physics and Astronomy Univ. of British Columbia Introduction Martin Weinberg reported

More information

Extreme Value Theory in (Hourly) Precipitation

Extreme Value Theory in (Hourly) Precipitation Extreme Value Theory in (Hourly) Precipitation Uli Schneider Geophysical Statistics Project, NCAR GSP Miniseries at CSU November 17, 2003 Outline Project overview Extreme value theory 101 Applying extreme

More information

Integration. Volume Estimation

Integration. Volume Estimation Monte Carlo Integration Lab Objective: Many important integrals cannot be evaluated symbolically because the integrand has no antiderivative. Traditional numerical integration techniques like Newton-Cotes

More information

Image Segmentation using Gaussian Mixture Models

Image Segmentation using Gaussian Mixture Models Image Segmentation using Gaussian Mixture Models Rahman Farnoosh, Gholamhossein Yari and Behnam Zarpak Department of Applied Mathematics, University of Science and Technology, 16844, Narmak,Tehran, Iran

More information

Passive Differential Matched-field Depth Estimation of Moving Acoustic Sources

Passive Differential Matched-field Depth Estimation of Moving Acoustic Sources Lincoln Laboratory ASAP-2001 Workshop Passive Differential Matched-field Depth Estimation of Moving Acoustic Sources Shawn Kraut and Jeffrey Krolik Duke University Department of Electrical and Computer

More information

Markov Random Fields and Gibbs Sampling for Image Denoising

Markov Random Fields and Gibbs Sampling for Image Denoising Markov Random Fields and Gibbs Sampling for Image Denoising Chang Yue Electrical Engineering Stanford University changyue@stanfoed.edu Abstract This project applies Gibbs Sampling based on different Markov

More information

Univariate Extreme Value Analysis. 1 Block Maxima. Practice problems using the extremes ( 2.0 5) package. 1. Pearson Type III distribution

Univariate Extreme Value Analysis. 1 Block Maxima. Practice problems using the extremes ( 2.0 5) package. 1. Pearson Type III distribution Univariate Extreme Value Analysis Practice problems using the extremes ( 2.0 5) package. 1 Block Maxima 1. Pearson Type III distribution (a) Simulate 100 maxima from samples of size 1000 from the gamma

More information

3 : Representation of Undirected GMs

3 : Representation of Undirected GMs 0-708: Probabilistic Graphical Models 0-708, Spring 202 3 : Representation of Undirected GMs Lecturer: Eric P. Xing Scribes: Nicole Rafidi, Kirstin Early Last Time In the last lecture, we discussed directed

More information