The Equivalence Theorem in Optimal Design

Σχετικά έγγραφα
Homework for 1/27 Due 2/5

Introduction of Numerical Analysis #03 TAGAMI, Daisuke (IMI, Kyushu University)

Lecture 17: Minimum Variance Unbiased (MVUB) Estimators

p n r

1. For each of the following power series, find the interval of convergence and the radius of convergence:

Outline. Detection Theory. Background. Background (Cont.)

LAD Estimation for Time Series Models With Finite and Infinite Variance

Lecture 3: Asymptotic Normality of M-estimators

true value θ. Fisher information is meaningful for families of distribution which are regular: W (x) f(x θ)dx

Μια εισαγωγή στα Μαθηματικά για Οικονομολόγους

HMY 795: Αναγνώριση Προτύπων

Χαρακτηριστικά της ανάλυσης διασποράς. ΑΝΑΛΥΣΗ ΙΑΣΠΟΡΑΣ (One-way analysis of variance)

Solutions: Homework 3

SUPERPOSITION, MEASUREMENT, NORMALIZATION, EXPECTATION VALUES. Reading: QM course packet Ch 5 up to 5.6

Variance Covariance Matrices for Linear Regression with Errors in both Variables. by J.W. Gillard and T.C. Iles

Probability theory STATISTICAL MODELING OF MULTIVARIATE EXTREMES, FMSN15/MASM23 TABLE OF FORMULÆ. Basic probability theory

Last Lecture. Biostatistics Statistical Inference Lecture 19 Likelihood Ratio Test. Example of Hypothesis Testing.

Parameter Estimation Fitting Probability Distributions Bayesian Approach

Bessel function for complex variable

Statistical Inference I Locally most powerful tests

Other Test Constructions: Likelihood Ratio & Bayes Tests

Solve the difference equation

Ψηφιακή Επεξεργασία Εικόνας

Latent variable models Variational approximations.

Αναγνώριση Προτύπων. Non Parametric

Diane Hu LDA for Audio Music April 12, 2010

1. Matrix Algebra and Linear Economic Models

Estimation for ARMA Processes with Stable Noise. Matt Calder & Richard A. Davis Colorado State University

Fourier Series. constant. The ;east value of T>0 is called the period of f(x). f(x) is well defined and single valued periodic function

HermiteHGeneral. Notations. Primary definition. Specific values. Traditional name. Traditional notation. Mathematica StandardForm notation

Bayesian statistics. DS GA 1002 Probability and Statistics for Data Science.

Factorial. Notations. Specific values. Traditional name. Traditional notation. Mathematica StandardForm notation. Specialized values

Latent variable models Variational approximations.

MEI EXAMINATION FORMULAE AND TABLES (MF2)

n r f ( n-r ) () x g () r () x (1.1) = Σ g() x = Σ n f < -n+ r> g () r -n + r dx r dx n + ( -n,m) dx -n n+1 1 -n -1 + ( -n,n+1)

α β

Estimators when the Correlation Coefficient. is Negative

Endogoneity and All That

DIPLOMA PROGRAMME MATHEMATICS SL INFORMATION BOOKLET

ΕΛΛΗΝΙΚΗ ΔΗΜΟΚΡΑΤΙΑ ΠΑΝΕΠΙΣΤΗΜΙΟ ΚΡΗΤΗΣ. Ψηφιακή Οικονομία. Διάλεξη 7η: Consumer Behavior Mαρίνα Μπιτσάκη Τμήμα Επιστήμης Υπολογιστών

J. of Math. (PRC) Shannon-McMillan, , McMillan [2] Breiman [3] , Algoet Cover [10] AEP. P (X n m = x n m) = p m,n (x n m) > 0, x i X, 0 m i n. (1.

Biorthogonal Wavelets and Filter Banks via PFFS. Multiresolution Analysis (MRA) subspaces V j, and wavelet subspaces W j. f X n f, τ n φ τ n φ.

CS 1675 Introduction to Machine Learning Lecture 7. Density estimation. Milos Hauskrecht 5329 Sennott Square

Outline. M/M/1 Queue (infinite buffer) M/M/1/N (finite buffer) Networks of M/M/1 Queues M/G/1 Priority Queue

FREE VIBRATION OF A SINGLE-DEGREE-OF-FREEDOM SYSTEM Revision B

CHAPTER 103 EVEN AND ODD FUNCTIONS AND HALF-RANGE FOURIER SERIES

Sixth Term Examination Papers MATHEMATICS LIST OF FORMULAE AND STATISTICAL TABLES

Exercises to Statistics of Material Fatigue No. 5

Presentation of complex number in Cartesian and polar coordinate system

6. MAXIMUM LIKELIHOOD ESTIMATION

Lecture 21: Properties and robustness of LSE


A NEW CLASS OF SEMI-PARAMETRIC ESTIMATORS OF THE SECOND ORDER PARAMETER *


Numerical Analysis FMN011

Tired Waiting in Queues? Then get in line now to learn more about Queuing!

ΚΕΦΑΛΑΙΟ 2 ΑΝΑΣΚΟΠΗΣΗ ΑΠΑΡΑΙΤΗΤΩΝ ΓΝΩΣΕΩΝ: ΕΚΤΙΜΗΤΕΣ

t-distribution t a (ν) s N μ = where X s s x = ν 2 FD ν 1 FD a/2 a/2 t-distribution normal distribution for ν>120

Generalized additive models in R

Three Classical Tests; Wald, LM(Score), and LR tests

9.1 Introduction 9.2 Lags in the Error Term: Autocorrelation 9.3 Estimating an AR(1) Error Model 9.4 Testing for Autocorrelation 9.

Dimension-free PAC-Bayesian bounds for matrices, vectors, and linear least squares regression.

Proof of Lemmas Lemma 1 Consider ξ nt = r

LIST OF FORMULAE STATISTICAL TABLES MATHEMATICS. (List MF1) AND

Supplemental Material to Comparison of inferential methods in partially identified models in terms of error in coverage probability

HMY 795: Αναγνώριση Προτύπων

A New Class of Analytic p-valent Functions with Negative Coefficients and Fractional Calculus Operators

Appendix B: Mathematical Formulae and Statistical Tables

IIT JEE (2013) (Trigonomtery 1) Solutions

INTEGRATION OF THE NORMAL DISTRIBUTION CURVE

Αλγόριθμοι και πολυπλοκότητα NP-Completeness (2)

Στοχαστικά Σήματα και Τηλεπικοινωνιές

Ψηφιακή Οικονομία. Διάλεξη 11η: Markets and Strategic Interaction in Networks Mαρίνα Μπιτσάκη Τμήμα Επιστήμης Υπολογιστών

ECE Spring Prof. David R. Jackson ECE Dept. Notes 2

Partial Differential Equations in Biology The boundary element method. March 26, 2013

The Heisenberg Uncertainty Principle

Exercises 10. Find a fundamental matrix of the given system of equations. Also find the fundamental matrix Φ(t) satisfying Φ(0) = I. 1.

상대론적고에너지중이온충돌에서 제트입자와관련된제동복사 박가영 인하대학교 윤진희교수님, 권민정교수님

Supplemental Material: Scaling Up Sparse Support Vector Machines by Simultaneous Feature and Sample Reduction

PARTIAL NOTES for 6.1 Trigonometric Identities

Fibonacci. Notations. Primary definition. Specific values. Traditional name. Traditional notation. Mathematica StandardForm notation

Matrices and Determinants

6.3 Forecasting ARMA processes

A Hierarchy of Theta Bodies for Polynomial Systems

ASYMPTOTIC BEST LINEAR UNBIASED ESTIMATION FOR THE LOG-GAMMA DISTRIBUTION

Στα επόμενα θεωρούμε ότι όλα συμβαίνουν σε ένα χώρο πιθανότητας ( Ω,,P) Modes of convergence: Οι τρόποι σύγκλισης μιας ακολουθίας τ.μ.

Degenerate Perturbation Theory

τατιςτική ςτην Εκπαίδευςη II

The Simply Typed Lambda Calculus

6.1. Dirac Equation. Hamiltonian. Dirac Eq.

Econ 2110: Fall 2008 Suggested Solutions to Problem Set 8 questions or comments to Dan Fetter 1

Nonlinear Motion. x M x. x x. cos. 2sin. tan. x x. Sextupoles cause nonlinear dynamics, which can be chaotic and unstable. CHESS & LEPP CHESS & LEPP

APPENDIX A DERIVATION OF JOINT FAILURE DENSITIES

Solution Series 9. i=1 x i and i=1 x i.

Chapter 7 Registers and Register Transfers

CHAPTER 101 FOURIER SERIES FOR PERIODIC FUNCTIONS OF PERIOD

ΠΑΝΕΠΙΣΤΗΜΙΟ ΚΥΠΡΟΥ ΤΜΗΜΑ ΠΛΗΡΟΦΟΡΙΚΗΣ. ΕΠΛ342: Βάσεις Δεδομένων. Χειμερινό Εξάμηνο Φροντιστήριο 10 ΛΥΣΕΙΣ. Επερωτήσεις SQL

Adaptive Covariance Estimation with model selection

Figure A.2: MPC and MPCP Age Profiles (estimating ρ, ρ = 2, φ = 0.03)..

CDMA. Performance Analysis of Chaotic Spread Spectrum CDMA Systems. LI Xiao - chao, GUO Dong - hui, ZENG Quan, WU Bo - xi RESEARCH & DEVELOPMENT

Transcript:

he Equivalece heorem i Optimal Desig Raier Schwabe & homas Schmelter, Otto vo Guericke Uiversity agdeburg Bayer Scherig Pharma, Berli rschwabe@ovgu.de PODE 007 ay 4, 007

Outlie Prologue: Simple eamples. A shortcut to liear models. Desigs ad desig criteria 3. Approimate desigs 4. he Equivalece heorem 5. RCR models Epilogue: Ope problems

Prologue: Simple eamples Compariso of two groups Y i μ + ε i i radom error observatio,..., i group i, group mea ε i i.i.d. Var ε σ i δ μ μ?

Compariso of two groups estimated differece δˆ y y Var δ ˆ + σ + fi optimal choice eve: / odd: ± /

Compariso of two groups optimal choice / 00 % 80 % deficiecy: 60 % 40 % icrease i sample size 0 % 0 % 33 % 5 % 50 % 0 0. 0.4 0.6 0.8 proportio: /

Liear regressio Y observatio,..., i Y 3.5.5 0.5 0-0.5 α + β + itercept slope ε eplaatory variable 0 0. 0.4 0.6 0.8 radom error ε i.i.d. Var ε σ

Liear regressio estimated slope ˆ β y y Var β ˆ σ desig regio: 0 optimal choice eve: 0 or, each / times

. A shortcut to liear models geeral liear model Y f β + ε radom error observatio,..., eplaatory variable ε i.i.d. Var ε σ regressio fuctios f f,..., f p β,..., β p β parameter

ε Fβ Y + Y Y Vector otatio f f F desig matri p I ε Y σ Cov Cov

Estimatio Gauss arkov heorem βˆ F F F Y least squares Is best liear ubiased esrimator for β covariace matri Cov βˆ F σ F iformatio matri F F Fisher iformatio f f

Y ε β β + + Liear regressio f f, regressio fuctios dimesio p F F iformatio matri

. Desigs ad desig criteria eact desig,..., iformatio matri,..., f f aim: choose,, m from desig regio X Covβˆ resp. to. miimise v f f predictio variace

Desig criteria miimise equivaletly l det D: A: det - trace - ISE: f - f d G: ma f - f

3. Approimate desigs... m... m desig poits weights iformatio matri f f D-optimal miimises det etc.

......,..., iformatio matri f f Embeddig of eact desigs stadardised

Coveity set Ξ of approimate desigs is cove α + α is a desig stadard criteria Φ : Ξ, ] are cove Φ α + α α Φ + α Φ D-criterio: Φ l det

4. he Equivalece heorem directioal derivative F ; η lim [ Φ Φ α α 0 α + α η at i the directio of η Φ ] ote: F Φ ; 0 regularity coditio: { f ; X} is compact

he Geeral Equivalece heorem miimises Φ if ad oly if Φ-opt. F Φ ; η 0 for all η maimises mi F η Φ ; η

Φ diffferetiable Sesitivity FΦ ; η η FΦ ; δ δ ϕ sesitivity fuctio ; F ; δ Φ miimises Φ ϕ ; 0 for all miimises maϕ ;

Sesitivity ϕ sesitivity fuctio ; F ; δ Φ miimises Φ ϕ ; 0 for all miimises maϕ ; ote: ϕ ; 0 for > 0

D-criterio l det Φ tr ; F - p η η Φ p - ; f f ϕ p - ; v f f D-optimal for all Kiefer, Wolfowitz 960 G-optimal

Liear criteria tr - A Φ tr tr ; F - - - η η A A Φ tr ; - - - ϕ A f A f tr - - f f A-optimal for all Fedorov 97: AI c-optimal c c c f - -

Pros ad cos usually ot costructive eceptio:polyomial regressio efficiecy bouds algorithms

Polyomial regressio Y, β + β +,, + β + ε p p 0 ϕ ; polyomial of degree p- miimal support 0 0 D-optimal:...... p / p p

Efficiecy bouds ; ma ϕ X + Φ Φ coveity ; F η η Φ Φ Φ D-criterio ; mav det det eff p p - - e D A-criterio tr ma eff A - - f f Φ Φ

Algorithms X arg maϕ ; steepest descet Fedorov Wy + α α α δ add oe poit α arg mi Φ α + αδ X

σ ukow iformatio matri for, σ β θ 4 0 0 σ σ θ 4 0 0 σ σ θ costat same story as before

5. Radom Coefficiet Regressio idividual curves are give by a commo liear model Y f b i i i + ε i idividual i,..., replicatio,...,m i idividual parameters: eplaatory variable populatio parameter b i ε ~ i error N 0, σ ~ N β, σ D idepedet

Desig idividual desig μ μ i i i i i...... populatio desig ν ν ζ ζ ζ i i i i...... problem: set of iformatio matrices is ot cove m i m

Sigle group desigs ζ iformatio for populatio parameter β + D m fied model iformatio

miimise D tr A + m Liear criteria m A tr AD tr + costat! A, ISE, c result Luoma 000, Liski et al. 00 optimal i reduced model optimal i RCR model

A Equivalece heorem coveity Φ Φ f cove cove i D-optimal for β - tr β locally optimal at D β β + D m - - f for all Fedorov, Hackl 997

Variace compoets coveity fails withi idividuals eve for sigle parameters e.g. d equivalece theorems which? give oly ecessary coditios

A ultivariate Equivalece heorem geeralised multivariate desigs ~,..., w~ m L L l,..., w~ l lm geeralised multivariate iformatio ~ ~ if ad oly if ~ sup m ~ l i w i tr V Fi Im + F ~ ~ det maimises,..., i ~ ~ F DF i ~ F i F settigs weights D-opt. p Fedorov 97

Epilogue: Ope Problems ubalaced desigs e.g. idividual desigs sigular sigle group desigs ot applicable for treatmet comparisos o-liear models liear approimatio: appropriate? Need for a more geeral approach!!!