Treatment and analysis of data Applied statistics Lecture 6: Bayesian estimation

Size: px
Start display at page:

Download "Treatment and analysis of data Applied statistics Lecture 6: Bayesian estimation"

Transcription

1 Treatment and analysis o data Applied statistics Lecture 6: Bayesian estimation Topics covered: Bayes' Theorem again Relation to Likelihood Transormation o pd A trivial example Wiener ilter Malmquist bias Lutz-Kelker bias Bayes versus Likelihood The bus problem Sept-Oct 2006 Statistics or astronomers (L. Lindegren, Lund Observatory) Lecture 6, p. 1 Bayesian estimation Thomas Bayes ( ) Thomas Bayes, British mathematician, Presbyterian minister and Fellow o the Royal Society. The manuscript Essay towards solving a problem in the doctrine o chances was ound ater his death and published It establishes a mathematical basis or probability inerence by: treating model parameters as random variables with a prior distribution prescribing how the distribution is modiied by data (Bayes theorem) basing the inerence on the resulting posterior distribution Sept-Oct 2006 Statistics or astronomers (L. Lindegren, Lund Observatory) Lecture 6, p. 2

2 Bayes' theorem P(A&B) = P(A)P(B A) = P(B)P(A B) A P(A B) = P(A)P(B A)/P(B) A = model (M), B = data (D) A&B P(M D) = P(M)P(D M)/P(D) B P(M) = prior probability o M (beore D) P(M D) = posterior probability o M (in light o D) P(D M) = likelihood o M (given D) P(D) = ixed [only needed to normalize P(M D)] Θ ( θ x ) = ( xθ ) ( xθ ) Θ Θ ( θ ) ( θ )dθ L( θ x ) Θ ( θ ) Sept-Oct 2006 Statistics or astronomers (L. Lindegren, Lund Observatory) Lecture 6, p. 3 Relation to Likelihood Θ ( θ x ) Θ( θ ) L( θ x ) posterior prior likelihood Treating θ as a random variable is still (ater 240 years!) a somewhat controversial issue. Also the choice o prior distribution is seen as problematic ( subjective ). I the prior pd is lat (over a reasonable interval o θ) then maximum a posteriori (MAP) is equivalent to maximum likelihood (ML). A slanted or peaked prior may push the MAP away rom the ML. I the data do not determine the parameter well (wide likelihood unction), then the posterior depends strongly on the prior. Conversely, or well-determined problems, the prior has little inluence. Note that Bayes theorem gives a pd or θ, not a value (point estimate). Sept-Oct 2006 Statistics or astronomers (L. Lindegren, Lund Observatory) Lecture 6, p. 4

3 Transormation o pd Let be a random variable with pd (x) and Y another random variable obtained by the transormation Y = g() where g is some known unction. What is the pd Y (y) o the transormed variable? The general case is complex, and Y (y) may not even exist. However, i g is continuous and monotone, the answer is simple: dx = Y ( y) dy Y ( y) = dx dy = g 1 Note that the is needed in case g is decreasing. Multivariate case: Y = g( ) Y ( y) = g det x T 1 The determinant is the Jacobian o the transormation. Sept-Oct 2006 Statistics or astronomers (L. Lindegren, Lund Observatory) Lecture 6, p. 5 A trivial (?) example (1/2) Suppose we want to measure the intensity λ o a source by counting the number o photons, n, detected in a certain time interval. We assume that n ~ Poisson(λ). Given n = 10, what is the estimate o λ? L(λ n) = λ n exp( λ) / n! => MLE λ* = n = 10 (reasonable!) Bayesian estimation (MAP) gives the same result i the prior distribution is lat between (say) 0 and 100. But i the prior state o knowledge is that we have no idea even about the order o magnitude o λ, then it can be argued that the prior pd is lat in log λ rather than λ. (C. requency o irst digit in natural constants!) This implies a prior pd inversely proportional to λ; thus posterior pd λ 1 L(λ n) = λ n 1 exp( λ) which has a maximum at λ = n 1. Thus, the Bayesian MAP estimate is 9. (This gets really weird when n = 1.) Sept-Oct 2006 Statistics or astronomers (L. Lindegren, Lund Observatory) Lecture 6, p. 6

4 A trivial (?) example (2/2) But is the MAP (maximum a posteriori) estimate really what we want? An alternative in Bayesian estimation is to compute the posterior mean. Using that λ n exp( λ) dλ = n! we ind or prior λ 0 : E(λ n) = n + 1 or prior λ 1 : E(λ n) = n The MAP estimate and the posterior mean are not invariant to transormation o λ (while the MLE is). There is yet another Bayesian estimate which is invariant to transormation, namely the posterior median. It is more complicated to compute, but to a good approximation we have or prior λ 0 : median(λ n) = n + 2/3 or prior λ 1 : median(λ n) = n 1/3 (n > 0) With dierent estimators we get anything between n 1 and n+1, but does it matter? Sept-Oct 2006 Statistics or astronomers (L. Lindegren, Lund Observatory) Lecture 6, p. 7 A more interesting example: Wiener ilter (1/2) Suppose we observe a continuous variable y = x + ε, where x and ε are independent Gaussian random variables with zero mean and s.d. s (or signal) and n (or noise): x ~ N(0, s 2 ), ε ~ N(0, n 2 ) Given the value y, what is the estimate o x? Prior pd or x : (x) = (2π) 1/2 s 1 exp[ x 2 /2s 2 ] Likelihood unction : L(x y) = (2π) 1/2 n 1 exp[ (y x) 2 /2n 2 ] Posterior pd or x : (x y) exp[ x 2 /2s 2 (y x) 2 /2n 2 ] 2 2 s n Completing the square, we ind that (x y) is Gaussian with variance = 2 2 s + n s and mean value xˆ = y, which is the Bayesian estimate o x. 2 2 s + n (A Wiener ilter has the transer unction R 2 /(1 + R 2 ), where R = S/N.) 2 Sept-Oct 2006 Statistics or astronomers (L. Lindegren, Lund Observatory) Lecture 6, p. 8

5 Wiener ilter (2/2) Sometimes it is helpul to visualize Bayes theorem by means o the joint pd o the parameter (x) and data (y): y = x+ε y = x observed y xˆ x equiprobability curves Sept-Oct 2006 Statistics or astronomers (L. Lindegren, Lund Observatory) Lecture 6, p. 9 Another example: Malmquist bias (1/3) I a class o objects has an intrinsic spread in luminosity (or absolute magnitude M), and we pick at random an object on the sky with apparent magnitude m, then that object is likely to be more luminous than is typical or the class. Malmquist (Lund Medd. Ser. II, No. 22, 1920) derived the required correction to the mean observed M as unction o the intrinsic spread in M and the observed distribution o m. In Bayesian terms, the eect can be understood as the dierence between the prior (intrinsic or true) distribution o M, and the posterior (apparent) distribution o M or given m. For a certain class o objects, assume or simplicity: 1. that the intrinsic luminosity unction is M ~ N(M 0, σ 2 ); 2. that the objects are on average uniormly distributed in space; 3. that there is no extinction. Let x = m M = 5 log (r/10 pc) denote the distance modulus. Sept-Oct 2006 Statistics or astronomers (L. Lindegren, Lund Observatory) Lecture 6, p. 10

6 Malmquist bias (2/3) Assumptions 2 and 3 imply that the distance modulus x = m M = 5 log (r/10 pc) has the (improper) pd (x) x. Thus: (m M) = (m M) (m M) = exp[ γ (m M) ] ( γ = 0.6 ln 10 = ) (M) exp[ (M M 0 ) 2 /2σ 2 ] (M m) exp[ γ (m M) ] exp[ (M M 0 ) 2 /2σ 2 ] thus (M m) ~ N(M 0 γσ 2, σ 2 ) = exp[ γ (m M) (M M 0 ) 2 /2σ 2 ] = exp[ γ (m M 0 ) + γ 2 σ 2 /2 ] exp[ (M M 0 + γσ 2 ) 2 /2σ 2 ] so the mean abs. mag. o the objects with apparent magnitude m is M = M 0 γσ 2 Sept-Oct 2006 Statistics or astronomers (L. Lindegren, Lund Observatory) Lecture 6, p. 11 Malmquist bias (3/3) m equiprobability curves x = constant M M 0 M Sept-Oct 2006 Statistics or astronomers (L. Lindegren, Lund Observatory) Lecture 6, p. 12

7 Yet another example: Lutz-Kelker bias (1/3) Let p 0 be the true parallax o a star and p the measured value. Assume that the measurement errors are Gaussian with zero mean and s.d. σ. Then, or any given star (with true parallax = p 0 ), P( p < p 0 p 0 ) = P( p > p 0 p 0 ) i.e., positive and negative errors are equally probable. Now consider instead any given measured parallax value p. Then, in general, P( p 0 < p p ) P( p 0 > p p ) that is, positive and negative errors are not equally probable! This may at irst seem paradoxical, but a single example may be enough to make the statement credible: it is possible to obtain a negative value o the measured parallax, in which case P( p 0 < p p ) = 0 and P( p 0 > p p ) = 1. Sept-Oct 2006 Statistics or astronomers (L. Lindegren, Lund Observatory) Lecture 6, p. 13 Lutz-Kelker bias (2/3) Lutz & Kelker (PASP 85, 573, 1973) discussed the use o trigonometric parallaxes or luminosity calibration and derived a systematic correction depending on the relative parallax error (σ/p). In a stellar sample selected according to a lower limit on the observed parallax, the sample mean parallax is systematically too large, because the random errors will scatter more stars into the volume (with positive errors) than out o it (with negative errors). The eect can be ormulated in Bayesian terms. Let us assume 1. that the observed parallax has the distribution p ~ N( p 0, σ 2 ) 2. that the number density (n, in pc 3 ) o stars o a given class decreases exponentially with the height z above the Galactic plane: n = n 0 exp( z /H), H = scale height 3. that there is no extinction. As unction o distance r = 1/p 0 we have n(r) = n 0 exp( βr), where β = sin b /H. Sept-Oct 2006 Statistics or astronomers (L. Lindegren, Lund Observatory) Lecture 6, p. 14

8 Lutz-Kelker bias (3/3) The number o stars in solid angle ω with distance r to r+dr is dn = ω r 2 n(r) dr. r = p 1 0 dr = p 2 0 dp 0 dn/dp 0 p 4 0 exp( β/p 0 ), thus: ( p 0 ) p 4 0 exp( β/p 0 ) (prior) ( p p 0 ) exp[ (p p 0 ) 2 /2σ 2 ] (likelihood) ( p 0 p ) p 4 0 exp( β/p 0 (p p 0 ) 2 /2σ 2 ] (posterior) Lp0 ( ) P( p0) 0.2 B( p0) 0.15 Lp0 ( ) P( p0) B( p0) p0 p = 10, σ = 1, β = p0 p = 10, σ = 2, β = 10 Sept-Oct 2006 Statistics or astronomers (L. Lindegren, Lund Observatory) Lecture 6, p. 15 Bayes versus Likelihood Let D 1 and D 2 be two independent data sets relevant to the same model M. Since the data sets are independent, the total likelihood is L(M D 1,D 2 ) = L(M D 1 ) L(M D 2 ) I we regard D 1 as representing the knowledge about M beore introducing D 2, we have, ater renormalization, essentially Bayes' theorem. The nice thing about Bayesian theory is that it gives a ramework or treating the prior and posterior knowledge on exactly the same ooting. The Bayesian approach also encourages us to think about the a priori assumptions in any experiment, which is probably a good thing. Acknowledgement: In this lecture I have made use o some good ideas rom Ned Wright's Journal Club Talk on Statistics, Sept-Oct 2006 Statistics or astronomers (L. Lindegren, Lund Observatory) Lecture 6, p. 16

Treatment and analysis of data Applied statistics Lecture 4: Estimation

Treatment and analysis of data Applied statistics Lecture 4: Estimation Treatment and analysis of data Applied statistics Lecture 4: Estimation Topics covered: Hierarchy of estimation methods Modelling of data The likelihood function The Maximum Likelihood Estimate (MLE) Confidence

More information

Treatment and analysis of data Applied statistics Lecture 5: More on Maximum Likelihood

Treatment and analysis of data Applied statistics Lecture 5: More on Maximum Likelihood Treatment and analysis of data Applied statistics Lecture 5: More on Maximum Likelihood Topics covered: The multivariate Gaussian Ordinary Least Squares (OLS) Generalized Least Squares (GLS) Linear and

More information

Parameter Estimation. William H. Jefferys University of Texas at Austin Parameter Estimation 7/26/05 1

Parameter Estimation. William H. Jefferys University of Texas at Austin Parameter Estimation 7/26/05 1 Parameter Estimation William H. Jefferys University of Texas at Austin bill@bayesrules.net Parameter Estimation 7/26/05 1 Elements of Inference Inference problems contain two indispensable elements: Data

More information

Chapter 3: Maximum-Likelihood & Bayesian Parameter Estimation (part 1)

Chapter 3: Maximum-Likelihood & Bayesian Parameter Estimation (part 1) HW 1 due today Parameter Estimation Biometrics CSE 190 Lecture 7 Today s lecture was on the blackboard. These slides are an alternative presentation of the material. CSE190, Winter10 CSE190, Winter10 Chapter

More information

Who was Bayes? Bayesian Phylogenetics. What is Bayes Theorem?

Who was Bayes? Bayesian Phylogenetics. What is Bayes Theorem? Who was Bayes? Bayesian Phylogenetics Bret Larget Departments of Botany and of Statistics University of Wisconsin Madison October 6, 2011 The Reverand Thomas Bayes was born in London in 1702. He was the

More information

Bayesian Phylogenetics

Bayesian Phylogenetics Bayesian Phylogenetics Bret Larget Departments of Botany and of Statistics University of Wisconsin Madison October 6, 2011 Bayesian Phylogenetics 1 / 27 Who was Bayes? The Reverand Thomas Bayes was born

More information

GAUSSIAN PROCESS REGRESSION

GAUSSIAN PROCESS REGRESSION GAUSSIAN PROCESS REGRESSION CSE 515T Spring 2015 1. BACKGROUND The kernel trick again... The Kernel Trick Consider again the linear regression model: y(x) = φ(x) w + ε, with prior p(w) = N (w; 0, Σ). The

More information

Statistical Tools and Techniques for Solar Astronomers

Statistical Tools and Techniques for Solar Astronomers Statistical Tools and Techniques for Solar Astronomers Alexander W Blocker Nathan Stein SolarStat 2012 Outline Outline 1 Introduction & Objectives 2 Statistical issues with astronomical data 3 Example:

More information

Lecture 13: Applications of Fourier transforms (Recipes, Chapter 13)

Lecture 13: Applications of Fourier transforms (Recipes, Chapter 13) Lecture 13: Applications o Fourier transorms (Recipes, Chapter 13 There are many applications o FT, some o which involve the convolution theorem (Recipes 13.1: The convolution o h(t and r(t is deined by

More information

Stat260: Bayesian Modeling and Inference Lecture Date: February 10th, Jeffreys priors. exp 1 ) p 2

Stat260: Bayesian Modeling and Inference Lecture Date: February 10th, Jeffreys priors. exp 1 ) p 2 Stat260: Bayesian Modeling and Inference Lecture Date: February 10th, 2010 Jeffreys priors Lecturer: Michael I. Jordan Scribe: Timothy Hunter 1 Priors for the multivariate Gaussian Consider a multivariate

More information

Parametric Techniques Lecture 3

Parametric Techniques Lecture 3 Parametric Techniques Lecture 3 Jason Corso SUNY at Buffalo 22 January 2009 J. Corso (SUNY at Buffalo) Parametric Techniques Lecture 3 22 January 2009 1 / 39 Introduction In Lecture 2, we learned how to

More information

Lecture 5. G. Cowan Lectures on Statistical Data Analysis Lecture 5 page 1

Lecture 5. G. Cowan Lectures on Statistical Data Analysis Lecture 5 page 1 Lecture 5 1 Probability (90 min.) Definition, Bayes theorem, probability densities and their properties, catalogue of pdfs, Monte Carlo 2 Statistical tests (90 min.) general concepts, test statistics,

More information

Digital Image Processing. Lecture 6 (Enhancement) Bu-Ali Sina University Computer Engineering Dep. Fall 2009

Digital Image Processing. Lecture 6 (Enhancement) Bu-Ali Sina University Computer Engineering Dep. Fall 2009 Digital Image Processing Lecture 6 (Enhancement) Bu-Ali Sina University Computer Engineering Dep. Fall 009 Outline Image Enhancement in Spatial Domain Spatial Filtering Smoothing Filters Median Filter

More information

1. Definition: Order Statistics of a sample.

1. Definition: Order Statistics of a sample. AMS570 Order Statistics 1. Deinition: Order Statistics o a sample. Let X1, X2,, be a random sample rom a population with p.d.. (x). Then, 2. p.d.. s or W.L.O.G.(W thout Loss o Ge er l ty), let s ssu e

More information

STAT 801: Mathematical Statistics. Hypothesis Testing

STAT 801: Mathematical Statistics. Hypothesis Testing STAT 801: Mathematical Statistics Hypothesis Testing Hypothesis testing: a statistical problem where you must choose, on the basis o data X, between two alternatives. We ormalize this as the problem o

More information

SYDE 372 Introduction to Pattern Recognition. Probability Measures for Classification: Part I

SYDE 372 Introduction to Pattern Recognition. Probability Measures for Classification: Part I SYDE 372 Introduction to Pattern Recognition Probability Measures for Classification: Part I Alexander Wong Department of Systems Design Engineering University of Waterloo Outline 1 2 3 4 Why use probability

More information

Naïve Bayes classification

Naïve Bayes classification Naïve Bayes classification 1 Probability theory Random variable: a variable whose possible values are numerical outcomes of a random phenomenon. Examples: A person s height, the outcome of a coin toss

More information

Machine Learning. Lecture 4: Regularization and Bayesian Statistics. Feng Li. https://funglee.github.io

Machine Learning. Lecture 4: Regularization and Bayesian Statistics. Feng Li. https://funglee.github.io Machine Learning Lecture 4: Regularization and Bayesian Statistics Feng Li fli@sdu.edu.cn https://funglee.github.io School of Computer Science and Technology Shandong University Fall 207 Overfitting Problem

More information

Universität Potsdam Institut für Informatik Lehrstuhl Maschinelles Lernen. Bayesian Learning. Tobias Scheffer, Niels Landwehr

Universität Potsdam Institut für Informatik Lehrstuhl Maschinelles Lernen. Bayesian Learning. Tobias Scheffer, Niels Landwehr Universität Potsdam Institut für Informatik Lehrstuhl Maschinelles Lernen Bayesian Learning Tobias Scheffer, Niels Landwehr Remember: Normal Distribution Distribution over x. Density function with parameters

More information

MAT Inverse Problems, Part 2: Statistical Inversion

MAT Inverse Problems, Part 2: Statistical Inversion MAT-62006 Inverse Problems, Part 2: Statistical Inversion S. Pursiainen Department of Mathematics, Tampere University of Technology Spring 2015 Overview The statistical inversion approach is based on the

More information

Hypothesis Testing. Econ 690. Purdue University. Justin L. Tobias (Purdue) Testing 1 / 33

Hypothesis Testing. Econ 690. Purdue University. Justin L. Tobias (Purdue) Testing 1 / 33 Hypothesis Testing Econ 690 Purdue University Justin L. Tobias (Purdue) Testing 1 / 33 Outline 1 Basic Testing Framework 2 Testing with HPD intervals 3 Example 4 Savage Dickey Density Ratio 5 Bartlett

More information

(a) B-V 6 V. (b) B-V

(a) B-V 6 V. (b) B-V 721 TOWARDS AN IMPROVED MODEL OF THE GALAXY Johan Holmberg 1, Chris Flynn 2, Lennart Lindegren 1 1 Lund Observatory, Box 43, SE-22100 Lund, Sweden 2 Tuorla Observatory, Vaisalantie 20, FI-21500 Piikkio,

More information

Lecture : Probabilistic Machine Learning

Lecture : Probabilistic Machine Learning Lecture : Probabilistic Machine Learning Riashat Islam Reasoning and Learning Lab McGill University September 11, 2018 ML : Many Methods with Many Links Modelling Views of Machine Learning Machine Learning

More information

Introduction to Bayesian Data Analysis

Introduction to Bayesian Data Analysis Introduction to Bayesian Data Analysis Phil Gregory University of British Columbia March 2010 Hardback (ISBN-10: 052184150X ISBN-13: 9780521841504) Resources and solutions This title has free Mathematica

More information

Estimation and detection of a periodic signal

Estimation and detection of a periodic signal Estimation and detection o a periodic signal Daniel Aronsson, Erik Björnemo, Mathias Johansson Signals and Systems Group, Uppsala University, Sweden, e-mail: Daniel.Aronsson,Erik.Bjornemo,Mathias.Johansson}@Angstrom.uu.se

More information

Supplement To: Search for Tensor, Vector, and Scalar Polarizations in the Stochastic Gravitational-Wave Background

Supplement To: Search for Tensor, Vector, and Scalar Polarizations in the Stochastic Gravitational-Wave Background Supplement To: Search or Tensor, Vector, and Scalar Polarizations in the Stochastic GravitationalWave Background B. P. Abbott et al. (LIGO Scientiic Collaboration & Virgo Collaboration) This documents

More information

Lecture Notes 1 Probability and Random Variables. Conditional Probability and Independence. Functions of a Random Variable

Lecture Notes 1 Probability and Random Variables. Conditional Probability and Independence. Functions of a Random Variable Lecture Notes 1 Probability and Random Variables Probability Spaces Conditional Probability and Independence Random Variables Functions of a Random Variable Generation of a Random Variable Jointly Distributed

More information

Lecture 3. G. Cowan. Lecture 3 page 1. Lectures on Statistical Data Analysis

Lecture 3. G. Cowan. Lecture 3 page 1. Lectures on Statistical Data Analysis Lecture 3 1 Probability (90 min.) Definition, Bayes theorem, probability densities and their properties, catalogue of pdfs, Monte Carlo 2 Statistical tests (90 min.) general concepts, test statistics,

More information

Lecture Notes 1 Probability and Random Variables. Conditional Probability and Independence. Functions of a Random Variable

Lecture Notes 1 Probability and Random Variables. Conditional Probability and Independence. Functions of a Random Variable Lecture Notes 1 Probability and Random Variables Probability Spaces Conditional Probability and Independence Random Variables Functions of a Random Variable Generation of a Random Variable Jointly Distributed

More information

Parametric Techniques

Parametric Techniques Parametric Techniques Jason J. Corso SUNY at Buffalo J. Corso (SUNY at Buffalo) Parametric Techniques 1 / 39 Introduction When covering Bayesian Decision Theory, we assumed the full probabilistic structure

More information

Naïve Bayes classification. p ij 11/15/16. Probability theory. Probability theory. Probability theory. X P (X = x i )=1 i. Marginal Probability

Naïve Bayes classification. p ij 11/15/16. Probability theory. Probability theory. Probability theory. X P (X = x i )=1 i. Marginal Probability Probability theory Naïve Bayes classification Random variable: a variable whose possible values are numerical outcomes of a random phenomenon. s: A person s height, the outcome of a coin toss Distinguish

More information

Bayesian Paradigm. Maximum A Posteriori Estimation

Bayesian Paradigm. Maximum A Posteriori Estimation Bayesian Paradigm Maximum A Posteriori Estimation Simple acquisition model noise + degradation Constraint minimization or Equivalent formulation Constraint minimization Lagrangian (unconstraint minimization)

More information

In many diverse fields physical data is collected or analysed as Fourier components.

In many diverse fields physical data is collected or analysed as Fourier components. 1. Fourier Methods In many diverse ields physical data is collected or analysed as Fourier components. In this section we briely discuss the mathematics o Fourier series and Fourier transorms. 1. Fourier

More information

Parametric Models. Dr. Shuang LIANG. School of Software Engineering TongJi University Fall, 2012

Parametric Models. Dr. Shuang LIANG. School of Software Engineering TongJi University Fall, 2012 Parametric Models Dr. Shuang LIANG School of Software Engineering TongJi University Fall, 2012 Today s Topics Maximum Likelihood Estimation Bayesian Density Estimation Today s Topics Maximum Likelihood

More information

ELEG 3143 Probability & Stochastic Process Ch. 4 Multiple Random Variables

ELEG 3143 Probability & Stochastic Process Ch. 4 Multiple Random Variables Department o Electrical Engineering University o Arkansas ELEG 3143 Probability & Stochastic Process Ch. 4 Multiple Random Variables Dr. Jingxian Wu wuj@uark.edu OUTLINE 2 Two discrete random variables

More information

Statistical Methods for Particle Physics Lecture 1: parameter estimation, statistical tests

Statistical Methods for Particle Physics Lecture 1: parameter estimation, statistical tests Statistical Methods for Particle Physics Lecture 1: parameter estimation, statistical tests http://benasque.org/2018tae/cgi-bin/talks/allprint.pl TAE 2018 Benasque, Spain 3-15 Sept 2018 Glen Cowan Physics

More information

Parameter estimation for nonlinear models: Numerical approaches to solving the inverse problem. Lecture 10 03/25/2008. Sven Zenker

Parameter estimation for nonlinear models: Numerical approaches to solving the inverse problem. Lecture 10 03/25/2008. Sven Zenker Parameter estimation or nonlinear models: Numerical approaches to solving the inverse problem Lecture 10 03/25/2008 Sven Zenker Review: Multiple Shooting homework Method o Multipliers: unction [x, lastlambda]

More information

Lecture 12: Distances to stars. Astronomy 111

Lecture 12: Distances to stars. Astronomy 111 Lecture 12: Distances to stars Astronomy 111 Why are distances important? Distances are necessary for estimating: Total energy released by an object (Luminosity) Masses of objects from orbital motions

More information

Figure 18: Top row: example of a purely continuous spectrum (left) and one realization

Figure 18: Top row: example of a purely continuous spectrum (left) and one realization 1..5 S(). -.2 -.5 -.25..25.5 64 128 64 128 16 32 requency time time Lag 1..5 S(). -.5-1. -.5 -.1.1.5 64 128 64 128 16 32 requency time time Lag Figure 18: Top row: example o a purely continuous spectrum

More information

Bayesian Regression Linear and Logistic Regression

Bayesian Regression Linear and Logistic Regression When we want more than point estimates Bayesian Regression Linear and Logistic Regression Nicole Beckage Ordinary Least Squares Regression and Lasso Regression return only point estimates But what if we

More information

Introduction to Bayesian Methods

Introduction to Bayesian Methods Introduction to Bayesian Methods Jessi Cisewski Department of Statistics Yale University Sagan Summer Workshop 2016 Our goal: introduction to Bayesian methods Likelihoods Priors: conjugate priors, non-informative

More information

SPOC: An Innovative Beamforming Method

SPOC: An Innovative Beamforming Method SPOC: An Innovative Beamorming Method Benjamin Shapo General Dynamics Ann Arbor, MI ben.shapo@gd-ais.com Roy Bethel The MITRE Corporation McLean, VA rbethel@mitre.org ABSTRACT The purpose o a radar or

More information

Additional exercises in Stationary Stochastic Processes

Additional exercises in Stationary Stochastic Processes Mathematical Statistics, Centre or Mathematical Sciences Lund University Additional exercises 8 * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * *

More information

STA414/2104 Statistical Methods for Machine Learning II

STA414/2104 Statistical Methods for Machine Learning II STA414/2104 Statistical Methods for Machine Learning II Murat A. Erdogdu & David Duvenaud Department of Computer Science Department of Statistical Sciences Lecture 3 Slide credits: Russ Salakhutdinov Announcements

More information

Markov Chain Monte Carlo

Markov Chain Monte Carlo Department of Statistics The University of Auckland https://www.stat.auckland.ac.nz/~brewer/ Emphasis I will try to emphasise the underlying ideas of the methods. I will not be teaching specific software

More information

Bayesian SAE using Complex Survey Data Lecture 4A: Hierarchical Spatial Bayes Modeling

Bayesian SAE using Complex Survey Data Lecture 4A: Hierarchical Spatial Bayes Modeling Bayesian SAE using Complex Survey Data Lecture 4A: Hierarchical Spatial Bayes Modeling Jon Wakefield Departments of Statistics and Biostatistics University of Washington 1 / 37 Lecture Content Motivation

More information

Physics 5153 Classical Mechanics. Solution by Quadrature-1

Physics 5153 Classical Mechanics. Solution by Quadrature-1 October 14, 003 11:47:49 1 Introduction Physics 5153 Classical Mechanics Solution by Quadrature In the previous lectures, we have reduced the number o eective degrees o reedom that are needed to solve

More information

Physics 403. Segev BenZvi. Choosing Priors and the Principle of Maximum Entropy. Department of Physics and Astronomy University of Rochester

Physics 403. Segev BenZvi. Choosing Priors and the Principle of Maximum Entropy. Department of Physics and Astronomy University of Rochester Physics 403 Choosing Priors and the Principle of Maximum Entropy Segev BenZvi Department of Physics and Astronomy University of Rochester Table of Contents 1 Review of Last Class Odds Ratio Occam Factors

More information

Gaussian Processes. Le Song. Machine Learning II: Advanced Topics CSE 8803ML, Spring 2012

Gaussian Processes. Le Song. Machine Learning II: Advanced Topics CSE 8803ML, Spring 2012 Gaussian Processes Le Song Machine Learning II: Advanced Topics CSE 8803ML, Spring 01 Pictorial view of embedding distribution Transform the entire distribution to expected features Feature space Feature

More information

7 Gaussian Discriminant Analysis (including QDA and LDA)

7 Gaussian Discriminant Analysis (including QDA and LDA) 36 Jonathan Richard Shewchuk 7 Gaussian Discriminant Analysis (including QDA and LDA) GAUSSIAN DISCRIMINANT ANALYSIS Fundamental assumption: each class comes from normal distribution (Gaussian). X N(µ,

More information

Measuring Statistical Evidence Using Relative Belief

Measuring Statistical Evidence Using Relative Belief Measuring Statistical Evidence Using Relative Belief Michael Evans Department of Statistics University of Toronto Abstract: A fundamental concern of a theory of statistical inference is how one should

More information

Bias in parallax measurements

Bias in parallax measurements Bias in parallax measurements When we measure some physical quantity experimentally, we usually incur some error in the measurement process, leading to an uncertainty in the result. Parallax is no exception:

More information

STATISTICS. 1. Measures of Central Tendency

STATISTICS. 1. Measures of Central Tendency STATISTICS 1. Measures o Central Tendency Mode, median and mean For a sample o discrete data, the mode is the observation, x with the highest requency,. 1 N F For grouped data in a cumulative requency

More information

Learning Bayesian network : Given structure and completely observed data

Learning Bayesian network : Given structure and completely observed data Learning Bayesian network : Given structure and completely observed data Probabilistic Graphical Models Sharif University of Technology Spring 2017 Soleymani Learning problem Target: true distribution

More information

Abstract. Statistics for Twenty-first Century Astrometry. Bayesian Analysis and Astronomy. Advantages of Bayesian Methods

Abstract. Statistics for Twenty-first Century Astrometry. Bayesian Analysis and Astronomy. Advantages of Bayesian Methods Abstract Statistics for Twenty-first Century Astrometry William H. Jefferys University of Texas at Austin, USA H.K. Eichhorn had a lively interest in statistics during his entire scientific career, and

More information

COM336: Neural Computing

COM336: Neural Computing COM336: Neural Computing http://www.dcs.shef.ac.uk/ sjr/com336/ Lecture 2: Density Estimation Steve Renals Department of Computer Science University of Sheffield Sheffield S1 4DP UK email: s.renals@dcs.shef.ac.uk

More information

The achievable limits of operational modal analysis. * Siu-Kui Au 1)

The achievable limits of operational modal analysis. * Siu-Kui Au 1) The achievable limits o operational modal analysis * Siu-Kui Au 1) 1) Center or Engineering Dynamics and Institute or Risk and Uncertainty, University o Liverpool, Liverpool L69 3GH, United Kingdom 1)

More information

CHAPTER 8 ANALYSIS OF AVERAGE SQUARED DIFFERENCE SURFACES

CHAPTER 8 ANALYSIS OF AVERAGE SQUARED DIFFERENCE SURFACES CAPTER 8 ANALYSS O AVERAGE SQUARED DERENCE SURACES n Chapters 5, 6, and 7, the Spectral it algorithm was used to estimate both scatterer size and total attenuation rom the backscattered waveorms by minimizing

More information

Density Estimation: ML, MAP, Bayesian estimation

Density Estimation: ML, MAP, Bayesian estimation Density Estimation: ML, MAP, Bayesian estimation CE-725: Statistical Pattern Recognition Sharif University of Technology Spring 2013 Soleymani Outline Introduction Maximum-Likelihood Estimation Maximum

More information

Bayesian Machine Learning

Bayesian Machine Learning Bayesian Machine Learning Andrew Gordon Wilson ORIE 6741 Lecture 4 Occam s Razor, Model Construction, and Directed Graphical Models https://people.orie.cornell.edu/andrew/orie6741 Cornell University September

More information

Structure and Evolution of Stars Lecture 13: Homology Solutions (#1)

Structure and Evolution of Stars Lecture 13: Homology Solutions (#1) Structure and Evolution o Stars Lecture 13: Homology Solutions (#1) Equations o Stellar Structure with mass as the independent variable Review o Observed relations between L, T and M Homology sel-similar

More information

Signal Detection and Estimation

Signal Detection and Estimation 394 based on the criteria o the unbiased and minimum variance estimator but rather on minimizing the squared dierence between the given data and the assumed signal data. We concluded the chapter with a

More information

Gaussian Process Regression Models for Predicting Stock Trends

Gaussian Process Regression Models for Predicting Stock Trends Gaussian Process Regression Models or Predicting Stock Trends M. Todd Farrell Andrew Correa December 5, 7 Introduction Historical stock price data is a massive amount o time-series data with little-to-no

More information

Overview. Probabilistic Interpretation of Linear Regression Maximum Likelihood Estimation Bayesian Estimation MAP Estimation

Overview. Probabilistic Interpretation of Linear Regression Maximum Likelihood Estimation Bayesian Estimation MAP Estimation Overview Probabilistic Interpretation of Linear Regression Maximum Likelihood Estimation Bayesian Estimation MAP Estimation Probabilistic Interpretation: Linear Regression Assume output y is generated

More information

7. Estimation and hypothesis testing. Objective. Recommended reading

7. Estimation and hypothesis testing. Objective. Recommended reading 7. Estimation and hypothesis testing Objective In this chapter, we show how the election of estimators can be represented as a decision problem. Secondly, we consider the problem of hypothesis testing

More information

Mixed Signal IC Design Notes set 6: Mathematics of Electrical Noise

Mixed Signal IC Design Notes set 6: Mathematics of Electrical Noise ECE45C /8C notes, M. odwell, copyrighted 007 Mied Signal IC Design Notes set 6: Mathematics o Electrical Noise Mark odwell University o Caliornia, Santa Barbara rodwell@ece.ucsb.edu 805-893-344, 805-893-36

More information

X-ray Diffraction. Interaction of Waves Reciprocal Lattice and Diffraction X-ray Scattering by Atoms The Integrated Intensity

X-ray Diffraction. Interaction of Waves Reciprocal Lattice and Diffraction X-ray Scattering by Atoms The Integrated Intensity X-ray Diraction Interaction o Waves Reciprocal Lattice and Diraction X-ray Scattering by Atoms The Integrated Intensity Basic Principles o Interaction o Waves Periodic waves characteristic: Frequency :

More information

Multivariate Bayesian Linear Regression MLAI Lecture 11

Multivariate Bayesian Linear Regression MLAI Lecture 11 Multivariate Bayesian Linear Regression MLAI Lecture 11 Neil D. Lawrence Department of Computer Science Sheffield University 21st October 2012 Outline Univariate Bayesian Linear Regression Multivariate

More information

MODULE 6 LECTURE NOTES 1 REVIEW OF PROBABILITY THEORY. Most water resources decision problems face the risk of uncertainty mainly because of the

MODULE 6 LECTURE NOTES 1 REVIEW OF PROBABILITY THEORY. Most water resources decision problems face the risk of uncertainty mainly because of the MODULE 6 LECTURE NOTES REVIEW OF PROBABILITY THEORY INTRODUCTION Most water resources decision problems ace the risk o uncertainty mainly because o the randomness o the variables that inluence the perormance

More information

DRAGON ADVANCED TRAINING COURSE IN ATMOSPHERE REMOTE SENSING. Inversion basics. Erkki Kyrölä Finnish Meteorological Institute

DRAGON ADVANCED TRAINING COURSE IN ATMOSPHERE REMOTE SENSING. Inversion basics. Erkki Kyrölä Finnish Meteorological Institute Inversion basics y = Kx + ε x ˆ = (K T K) 1 K T y Erkki Kyrölä Finnish Meteorological Institute Day 3 Lecture 1 Retrieval techniques - Erkki Kyrölä 1 Contents 1. Introduction: Measurements, models, inversion

More information

Statistical Methods for Astronomy

Statistical Methods for Astronomy Statistical Methods for Astronomy If your experiment needs statistics, you ought to have done a better experiment. -Ernest Rutherford Lecture 1 Lecture 2 Why do we need statistics? Definitions Statistical

More information

Probability and Estimation. Alan Moses

Probability and Estimation. Alan Moses Probability and Estimation Alan Moses Random variables and probability A random variable is like a variable in algebra (e.g., y=e x ), but where at least part of the variability is taken to be stochastic.

More information

Introduction to Machine Learning. Maximum Likelihood and Bayesian Inference. Lecturers: Eran Halperin, Yishay Mansour, Lior Wolf

Introduction to Machine Learning. Maximum Likelihood and Bayesian Inference. Lecturers: Eran Halperin, Yishay Mansour, Lior Wolf 1 Introduction to Machine Learning Maximum Likelihood and Bayesian Inference Lecturers: Eran Halperin, Yishay Mansour, Lior Wolf 2013-14 We know that X ~ B(n,p), but we do not know p. We get a random sample

More information

Basic concepts in estimation

Basic concepts in estimation Basic concepts in estimation Random and nonrandom parameters Definitions of estimates ML Maimum Lielihood MAP Maimum A Posteriori LS Least Squares MMS Minimum Mean square rror Measures of quality of estimates

More information

Choosing among models

Choosing among models Eco 515 Fall 2014 Chris Sims Choosing among models September 18, 2014 c 2014 by Christopher A. Sims. This document is licensed under the Creative Commons Attribution-NonCommercial-ShareAlike 3.0 Unported

More information

Frequentist-Bayesian Model Comparisons: A Simple Example

Frequentist-Bayesian Model Comparisons: A Simple Example Frequentist-Bayesian Model Comparisons: A Simple Example Consider data that consist of a signal y with additive noise: Data vector (N elements): D = y + n The additive noise n has zero mean and diagonal

More information

Lecture 6: Markov Chain Monte Carlo

Lecture 6: Markov Chain Monte Carlo Lecture 6: Markov Chain Monte Carlo D. Jason Koskinen koskinen@nbi.ku.dk Photo by Howard Jackman University of Copenhagen Advanced Methods in Applied Statistics Feb - Apr 2016 Niels Bohr Institute 2 Outline

More information

Parameter Estimation. Industrial AI Lab.

Parameter Estimation. Industrial AI Lab. Parameter Estimation Industrial AI Lab. Generative Model X Y w y = ω T x + ε ε~n(0, σ 2 ) σ 2 2 Maximum Likelihood Estimation (MLE) Estimate parameters θ ω, σ 2 given a generative model Given observed

More information

Transdimensional Markov Chain Monte Carlo Methods. Jesse Kolb, Vedran Lekić (Univ. of MD) Supervisor: Kris Innanen

Transdimensional Markov Chain Monte Carlo Methods. Jesse Kolb, Vedran Lekić (Univ. of MD) Supervisor: Kris Innanen Transdimensional Markov Chain Monte Carlo Methods Jesse Kolb, Vedran Lekić (Univ. of MD) Supervisor: Kris Innanen Motivation for Different Inversion Technique Inversion techniques typically provide a single

More information

Naïve Bayes. Jia-Bin Huang. Virginia Tech Spring 2019 ECE-5424G / CS-5824

Naïve Bayes. Jia-Bin Huang. Virginia Tech Spring 2019 ECE-5424G / CS-5824 Naïve Bayes Jia-Bin Huang ECE-5424G / CS-5824 Virginia Tech Spring 2019 Administrative HW 1 out today. Please start early! Office hours Chen: Wed 4pm-5pm Shih-Yang: Fri 3pm-4pm Location: Whittemore 266

More information

Hypothesis Testing, Bayes Theorem, & Parameter Estimation

Hypothesis Testing, Bayes Theorem, & Parameter Estimation Data Mining In Modern Astronomy Sky Surveys: Hypothesis Testing, Bayes Theorem, & Parameter Estimation Ching-Wa Yip cwyip@pha.jhu.edu; Bloomberg 518 Erratum of Last Lecture The Central Limit Theorem was

More information

CSC321 Lecture 18: Learning Probabilistic Models

CSC321 Lecture 18: Learning Probabilistic Models CSC321 Lecture 18: Learning Probabilistic Models Roger Grosse Roger Grosse CSC321 Lecture 18: Learning Probabilistic Models 1 / 25 Overview So far in this course: mainly supervised learning Language modeling

More information

Bayesian Inference in Astronomy & Astrophysics A Short Course

Bayesian Inference in Astronomy & Astrophysics A Short Course Bayesian Inference in Astronomy & Astrophysics A Short Course Tom Loredo Dept. of Astronomy, Cornell University p.1/37 Five Lectures Overview of Bayesian Inference From Gaussians to Periodograms Learning

More information

Physics 6720 Introduction to Statistics April 4, 2017

Physics 6720 Introduction to Statistics April 4, 2017 Physics 6720 Introduction to Statistics April 4, 2017 1 Statistics of Counting Often an experiment yields a result that can be classified according to a set of discrete events, giving rise to an integer

More information

Predictive Hypothesis Identification

Predictive Hypothesis Identification Marcus Hutter - 1 - Predictive Hypothesis Identification Predictive Hypothesis Identification Marcus Hutter Canberra, ACT, 0200, Australia http://www.hutter1.net/ ANU RSISE NICTA Marcus Hutter - 2 - Predictive

More information

Parameter estimation and forecasting. Cristiano Porciani AIfA, Uni-Bonn

Parameter estimation and forecasting. Cristiano Porciani AIfA, Uni-Bonn Parameter estimation and forecasting Cristiano Porciani AIfA, Uni-Bonn Questions? C. Porciani Estimation & forecasting 2 Temperature fluctuations Variance at multipole l (angle ~180o/l) C. Porciani Estimation

More information

Theory of Maximum Likelihood Estimation. Konstantin Kashin

Theory of Maximum Likelihood Estimation. Konstantin Kashin Gov 2001 Section 5: Theory of Maximum Likelihood Estimation Konstantin Kashin February 28, 2013 Outline Introduction Likelihood Examples of MLE Variance of MLE Asymptotic Properties What is Statistical

More information

2 Statistical Estimation: Basic Concepts

2 Statistical Estimation: Basic Concepts Technion Israel Institute of Technology, Department of Electrical Engineering Estimation and Identification in Dynamical Systems (048825) Lecture Notes, Fall 2009, Prof. N. Shimkin 2 Statistical Estimation:

More information

Doing Bayesian Integrals

Doing Bayesian Integrals ASTR509-13 Doing Bayesian Integrals The Reverend Thomas Bayes (c.1702 1761) Philosopher, theologian, mathematician Presbyterian (non-conformist) minister Tunbridge Wells, UK Elected FRS, perhaps due to

More information

Classical and Bayesian inference

Classical and Bayesian inference Classical and Bayesian inference AMS 132 Claudia Wehrhahn (UCSC) Classical and Bayesian inference January 8 1 / 11 The Prior Distribution Definition Suppose that one has a statistical model with parameter

More information

Statistical Methods in Particle Physics Lecture 1: Bayesian methods

Statistical Methods in Particle Physics Lecture 1: Bayesian methods Statistical Methods in Particle Physics Lecture 1: Bayesian methods SUSSP65 St Andrews 16 29 August 2009 Glen Cowan Physics Department Royal Holloway, University of London g.cowan@rhul.ac.uk www.pp.rhul.ac.uk/~cowan

More information

DD Advanced Machine Learning

DD Advanced Machine Learning Modelling Carl Henrik {chek}@csc.kth.se Royal Institute of Technology November 4, 2015 Who do I think you are? Mathematically competent linear algebra multivariate calculus Ok programmers Able to extend

More information

TLT-5200/5206 COMMUNICATION THEORY, Exercise 3, Fall TLT-5200/5206 COMMUNICATION THEORY, Exercise 3, Fall Problem 1.

TLT-5200/5206 COMMUNICATION THEORY, Exercise 3, Fall TLT-5200/5206 COMMUNICATION THEORY, Exercise 3, Fall Problem 1. TLT-5/56 COMMUNICATION THEORY, Exercise 3, Fall Problem. The "random walk" was modelled as a random sequence [ n] where W[i] are binary i.i.d. random variables with P[W[i] = s] = p (orward step with probability

More information

Physics 509: Error Propagation, and the Meaning of Error Bars. Scott Oser Lecture #10

Physics 509: Error Propagation, and the Meaning of Error Bars. Scott Oser Lecture #10 Physics 509: Error Propagation, and the Meaning of Error Bars Scott Oser Lecture #10 1 What is an error bar? Someone hands you a plot like this. What do the error bars indicate? Answer: you can never be

More information

CS4705. Probability Review and Naïve Bayes. Slides from Dragomir Radev

CS4705. Probability Review and Naïve Bayes. Slides from Dragomir Radev CS4705 Probability Review and Naïve Bayes Slides from Dragomir Radev Classification using a Generative Approach Previously on NLP discriminative models P C D here is a line with all the social media posts

More information

Winter 2019 Math 106 Topics in Applied Mathematics. Lecture 8: Importance Sampling

Winter 2019 Math 106 Topics in Applied Mathematics. Lecture 8: Importance Sampling Winter 2019 Math 106 Topics in Applied Mathematics Data-driven Uncertainty Quantification Yoonsang Lee (yoonsang.lee@dartmouth.edu) Lecture 8: Importance Sampling 8.1 Importance Sampling Importance sampling

More information

DS-GA 1003: Machine Learning and Computational Statistics Homework 7: Bayesian Modeling

DS-GA 1003: Machine Learning and Computational Statistics Homework 7: Bayesian Modeling DS-GA 1003: Machine Learning and Computational Statistics Homework 7: Bayesian Modeling Due: Tuesday, May 10, 2016, at 6pm (Submit via NYU Classes) Instructions: Your answers to the questions below, including

More information

Stat 5101 Lecture Notes

Stat 5101 Lecture Notes Stat 5101 Lecture Notes Charles J. Geyer Copyright 1998, 1999, 2000, 2001 by Charles J. Geyer May 7, 2001 ii Stat 5101 (Geyer) Course Notes Contents 1 Random Variables and Change of Variables 1 1.1 Random

More information

Physics 2B Chapter 17 Notes - First Law of Thermo Spring 2018

Physics 2B Chapter 17 Notes - First Law of Thermo Spring 2018 Internal Energy o a Gas Work Done by a Gas Special Processes The First Law o Thermodynamics p Diagrams The First Law o Thermodynamics is all about the energy o a gas: how much energy does the gas possess,

More information

Introduction to Probabilistic Machine Learning

Introduction to Probabilistic Machine Learning Introduction to Probabilistic Machine Learning Piyush Rai Dept. of CSE, IIT Kanpur (Mini-course 1) Nov 03, 2015 Piyush Rai (IIT Kanpur) Introduction to Probabilistic Machine Learning 1 Machine Learning

More information