Proceedings of COMPSTAT'2010 (eBook)

Preface 5
Acknowledgements 8
Sponsors 10
Contents 11
Part IKeynote 29
Complexity Questions in Non-Uniform Random Variate Generation 30
1 The pioneers 30
2 The assumptions and the limitations 32
3 The rejection method 37
4 The alternating series method 39
5 Oracles 41
6 Open questions 41
References 43
Computational Statistics Solutions for Molecular Biomedical Research: A Challenge and Chance for Both 46
1 Introduction 46
2 Screening Molecular Data for Predictors 48
2.1 Using the Analysis of Molecular Data for IdentifyingPredictive Biomarker 49
2.2 Combining the Analysis of Molecular Data for Prognosis 51
3 Outweighing Flexibility and Complexity Using Adaptive Designs 53
4 Discussion 56
5 First Author’s epilogue 57
References 58
The Laws of Coincidence 60
1 Introduction 60
2 Chance and coincidence 62
3 The Laws of coincidence 64
4 Conclusion 69
References 70
Part IIABC Methods for Genetic Data 71
Choosing the Summary Statistics and the Acceptance Rate in Approximate Bayesian Computation 72
1 Introduction 72
1.1 Rejection algorithm 72
1.2 Regression adjustment 73
1.3 Potential pitfalls of ABC 74
1.4 Outline of the paper 75
2 Regression adjustment in a Bayesian fashion 75
2.1 Local Bayesian regression 75
2.2 The evidence approximation 76
3 The evidence function as an omnibus criterion 77
3.1 Choosing the acceptance rate 77
3.2 Choosing the summary statistics 77
3.3 Choosing the scale of the summary statistics 79
3.4 Using the evidence without regression adjustment 80
References 81
Integrating Approximate Bayesian Computation with Complex Agent-Based Models for Cancer Research 82
1 Introduction 82
1.1 Agent-based modeling in cancer research 82
1.2 Coupling biological data and models with ABC 83
2 Material and methods 83
2.1 Methylation data 83
2.2 Modeling the colon crypt 84
2.3 Inferring colon crypt dynamics with ABC 87
3 Results 88
4 Discussion 89
References 90
Part IIIAlgorithms for Robust Statistics 92
Robust Model Selection with LARS Based on S-estimators 93
1 Introduction 93
2 Review of Least Angle Regression 95
3 LARS based on S-estimators 97
4 Simulation results 98
5 Conclusion 99
References 101
Robust Methods for Compositional Data 103
1 Compositional data and logratio transformations 103
2 Robustness for compositional data 106
2.1 Multivariate outlier detection 106
2.2 Principal component analysis (PCA) 107
2.3 Factor analysis 108
3 Real data example 108
4 Conclusions 110
References 112
Detecting Multivariate Outliers Using Projection Pursuit with Particle Swarm Optimization 113
1 Introduction 113
2 Projection indices for detecting outliers 114
3 Different “pursuit” strategies 116
4 Tribes: a parameter-free Particle Swarm optimization algorithm 117
5 Perspectives 118
Acknowledgements 120
References 121
Part IVBrain Imaging 123
Imaging Genetics: Bio-Informatics and Bio-Statistics Challenges 124
1 Neuroimaging genetics and the IMAGEN project 124
1.1 Genetic data: Single Nucleotide Polymorphisms (SNP) 125
1.2 Magnetic Resonance Imaging (MRI) data 125
2 Biostatistics: challenges and methods 128
2.1 Mappings one to many Voxel based mappings: BWAS. 128
2.2 Two-blocks methods 130
2.3 Multi block analyses: RGCCA 131
2.4 Biostatistics challenges and strategies for data analysis 132
3 Conclusions 132
Acknowledgements 132
References 133
The NPAIRS Computational Statistics Framework for Data Analysis in Neuroimaging 134
1 Introduction 134
2 Data-driven performance metrics 134
3 Nonparametric, activation, influence and reproducibility resampling (NPAIRS) 135
4 Measuring pipeline performance 139
5 Measuring dimensionality 140
References 142
Part VComputational Econometrics 144
Bootstrap Prediction in Unobserved Component Models 145
1 Introduction 145
2 The random walk plus noise model and the Kalman filter 146
3 PMSE of unobserved components 148
4 Prediction intervals of future observations 149
5 Empirical applications 151
5.1 Estimation of the NAIRU 151
5.2 Prediction of mortgage changes 152
6 Conclusions 152
References 153
Part VIComputer-Intensive Actuarial Methods 154
A Numerical Approach to Ruin Models with Excess of Loss Reinsurance and Reinstatements 155
1 Introduction 155
2 XL reinsurance with reinstatements 156
3 A recursive algorithm for the numerical solution 158
4 Numerical illustrations 159
References 163
Computation of the Aggregate Claim Amount Distribution Using R and Actuar 165
1 Introduction 165
2 The collective risk model 166
3 Discretization of claim amount distributions 166
4 Numerical evaluation of the aggregate claim amount distribution 168
4.1 Recursive calculation 168
4.2 Simulation 169
4.3 Direct calculation 169
4.4 Approximating distributions 169
5 Interface 170
6 Summary methods 171
7 Conclusion 173
Acknowledgments 174
References 174
Applications of Multilevel Structured Additive Regression Models to Insurance Data 175
1 Introduction 175
2 Priors for the regression coefficient 177
2.1 General form of basic priors 177
2.2 Compound priors 178
3 Sketch of MCMC Inference 178
3.1 Full conditionals for regression coefficients of nonlinear terms 179
3.2 Alternative sampling scheme based on a transformed parametrization 180
4 Applications to insurance data 181
4.1 Car insurance data 181
4.2 Health insurance data 182
References 184
Part VIIData Stream Mining 185
Temporally-Adaptive Linear Classification for Handling Population Drift in Credit Scoring 186
1 Introduction 186
2 Classification and credit scoring 187
3 Adaptive Linear Classifier 188
3.2 Building adaptive classifiers 190
4 Data and Experiments 191
5 Conclusion 195
Acknowledgements: 195
References 195
Large-Scale Machine Learning with Stochastic Gradient Descent 196
1 Introduction 196
2 Learning with gradient descent 196
2.1 Gradient descent 197
2.2 Stochastic gradient descent 197
2.3 Stochastic gradient examples 198
3 Learning with large training sets 199
3.1 The tradeoffs of large scale learning 199
3.2 Asymptotic analysis 200
4 Efficient learning 202
5 Experiments 204
References 204
Part VIIIFunctional Data Analysis 206
Anticipated and Adaptive Prediction in Functional Discriminant Analysis 207
1 Introduction 207
2 Linear discriminant analysis on functional data. The PLS approach 209
2.1 The PLS approximation 210
2.2 Quality criterion. The ROC curve 210
3 Anticipated and adaptive prediction 211
3.1 Anticipated prediction 211
3.2 Adaptive prediction 211
Adaptive prediction rule 212
4 Application 213
5 Conclusions 214
References 215
Bootstrap Calibration in Functional Linear Regression Models with Applications 217
1 Introduction 217
2 FPCA-type estimates 219
3 Bootstrap calibration 220
3.1 Building confidence intervals for prediction 220
3.2 Testing for lack of dependence 221
3.3 Testing for equality of model parameters 222
4 Simulation study 222
References 224
Empirical Dynamics and Functional Data Analysis 226
1 Introduction 226
2 Estimating derivatives from sparsely sampled data 228
3 The PACE Package 229
4 Empirical dynamics 232
Acknowledgments 233
References 233
Part IXKernel Methods 236
Indefinite Kernel Discriminant Analysis 237
1 Introduction 237
2 Kernels and Feature Space Embedding 238
3 Kernel Discriminant Analysis 239
3.1 Indefinite Kernel Fisher Discriminant Analysis 239
3.2 Indefinite Kernel Mahalanobis Distances 241
4 Classification Experiments 242
5 Conclusion 245
6 Acknowledgements 245
References 245
Data Dependent Priors in PAC-Bayes Bounds 247
1 Introduction 247
2 PAC-Bayes bound for SVM 248
3 Stretched Prior PAC-Bayes Bound 251
4 -Prior SVM 252
5 Experimental Work 253
6 Concluding remarks 254
Acknowledgments 256
References 256
Part XMonte Carlo Methods in System Safety, Reliability and Risk Analysis 257
Some Algorithms to Fit some Reliability Mixture Models under Censoring 258
1 The latent data model 258
Remark: 260
2 Parametric Stochastic-EM algorithm for model (1) 260
3 Nonparametric estimation under censoring 261
4 Semiparametric mixture models 262
4.1 Identifiability 262
4.2 Semiparametric Stochastic-EM algorithm for model (3) Without censoring. 262
5 Illustrative examples 264
5.1 A parametric example 264
5.2 A semiparametric example 264
References 264
Computational and Monte-Carlo Aspects of Systems for Monitoring Reliability Data 268
1 Introduction 268
2 Basic Approach 270
3 Computational and Monte-Carlo Aspects 272
4 Monitoring of Wearout 275
5 Conclusions 276
References 277
Part XIOptimization Heuristics in Statistical Modelling 278
Evolutionary Computation for Modelling and Optimization in Finance 279
1 Introduction 279
2 Evolutionary Computation 280
3 Clustering in Credit Risk Bucketing 282
4 Financial Portfolio Optimization 283
5 Conclusion and Further Research 286
6 Acknowledgement 287
References 287
Part XIISpatial Statistics / Spatial Epidemiology 289
Examining the Association between Deprivation Profiles and Air Pollution in Greater London using Bayesian Dirichlet Process Mixt 290
1 Introduction 290
2 Materials and Methods 291
2.1 Deprivation Profile Assignment Sub-Model 291
2.2 Deprivation Sub-Model 292
2.3 Finding the Clustering That Best Fits the Data 293
3 Example Association between Deprivation Profiles and Air Pollution Exposure 293
References 296
Assessing the Association between Environmental Exposures and Human Health 297
1 Introduction 297
2 The Study and Supporting Data 298
3 Change of Support 299
4 Spatial regression models with misaligned data 300
4.1 Traditional Krige and Regress (KR) 303
4.2 Traditional Krige and Regress with a General Covariance Structure (KRGC) 303
5 Simulation Study 304
6 Discussion 305
References 306
Part XIIIARS Session (Financial) Time Series 307
Semiparametric Seasonal Cointegrating Rank Selection 308
1 Introduction 308
2 The semiparametric seasonal ECM 309
3 Asymptotic results 310
4 Monte Carlo simulations 311
5 Conclusions 314
References 314
Estimating Factor Models for Multivariate Volatilities: An Innovation Expansion Method 316
1 Introduction 316
2 Models and methodology 317
3 Consistency of the estimator 320
4 Numerical properties 320
4.1 Simulated examples 321
4.2 A real data example 323
References 325
Multivariate Stochastic Volatility Model with Cross Leverage 326
1 Introduction 326
2 MSV model with cross leverage 327
2.2 MCMC implementation 328
3 Conclusion 332
Acknowledgement 332
A Derivation of the approximating state space model 333
References 333
Part XIVKDD Session: Topological Learning 335
Bag of Pursuits and Neural Gas for Improved Sparse Coding 336
1 Introduction 336
2 From vector quantization to sparse coding 337
3 A bag of orthogonal matching pursuits 339
4 Experiments 341
5 Conclusion 345
References 345
On the Role and Impact of the Metaparameters in t-distributed Stochastic Neighbor Embedding 346
1 Introduction 346
2 Stochastic Neighbor Embedding 347
3 Connection between similarity and distance preservation 349
4 Experiments 350
5 Discussion 351
6 Conclusions 354
References 354
Part XVIFCS Session: New Developments in Two or Highermode Clustering Model Based Clustering and Reduction for High Dimensional Data356
Multiple Nested Reductions of Single Data Modes as a Tool to Deal with Large Data Sets 357
1 Introduction 357
2 Principles 359
3 Examples 361
3.1 Existing models 361
3.2 Novel extension of existing model 362
4 Discussion 364
Acknowledgement footnote 366
References 366
The Generic Subspace Clustering Model 367
1 Introduction 367
2 Generic subspace clustering model 368
2.1 Loss function of the generic subspace clustering model 369
3 Positioning of existing subspace clustering approaches into the generic framework 370
3.1 Model variants: Reduced space at the between-level or within-level 370
3.2 Model variants: Common and distinctive models of the within-residuals 371
4 Discussion and conclusion 373
References 374
Clustering Discrete Choice Data 376
1 Motivation 376
2 The Model 377
3 ML Parameter Estimation 380
3.1 The EM algorithm 381
4 Discrete Choice Count Data 382
5 Example: crackers data 383
6 Concluding remarks 385
References 385
Part XVISelected Contributed Papers 386
Application of Local Influence Diagnostics to the Buckley-James Model 387
1 Introduction 387
2 Local Influence Diagnostics for the Buckley-James model 388
2.1 Perturbing the variance for censored regression 390
2.2 Perturbing response variables for censored regression 391
2.3 Perturbing independent variables for censored regression 391
3 Illustration 393
4 Conclusion 394
References 394
Multiblock Method for Categorical Variables. Application to the Study of Antibiotic Resistance 395
1 Introduction 395
2 Method 396
3 Application 399
3.1 Epidemiological data and objectives 399
3.2 Selection of the optimal models 399
3.3 Risk factors obtained from 399
3.4 Method comparison 400
4 Concluding remarks 400
References 402
A Flexible IRT Model for Health Questionnaire: an Application to HRQoL 403
1 Introduction 403
2 Material and methods 405
3 Results 408
4 Conclusion 409
References 410
Multidimensional Exploratory Analysis of a Structural Model Using a Class of Generalized Covariance Criteria 411
Introduction 411
1 Model and problem: 412
1.1 Thematic model 412
1.2 Problem 412
2 Thematic equation model exploration 414
2.1 Multiple covariance. 414
2.2 Multiple co-structure 414
2.3 THEME criterion for rank 1 components. 414
2.4 Beyond 1 component per group. 414
2.5 The local nesting principle. 415
2.6 Generic form of the criterion 415
2.7 Equivalent unconstrained minimization program. 416
S. 2.8 An alternative algorithm. 416
3 Chemometrical application 417
4 Conclusion and perspectives: 418
References 418
Semiparametric Models with Functional Responses in a Model Assisted Survey Sampling Setting : Model Assisted Estimation of Elect 419
1 Introduction 419
2 Functional data in a finite population 420
3 Semiparametric estimation with auxiliary information 422
4 Estimation of electricity consumption curves 423
Acknowledgment. 425
References 425
Stochastic Approximation for Multivariate and Functional Median 427
1 Introduction 427
2 A stochastic algorithm for online estimation of the median 429
3 A simulation study 430
4 Estimation of the median electricity consumption curve 431
References 433
A Markov Switching Re-evaluation of Event-Study Methodology 435
1 Introduction 435
2 The revised event-study methodology 437
3 Computational implementation 438
4 An Application 439
5 Conclusions 441
References 442
Evaluation of DNA Mixtures Accounting for Sampling Variability 443
1 Introduction 443
2 Likelihood ratio 444
3 Case Study 446
4 Discussion 448
Appendix 449
References 450
Monotone Graphical Multivariate Markov Chains 451
1 Introduction 451
2 Basic notation 452
3 Monotone and graphical multivariate Markov chains 452
4 A multivariate logistic model for transition probabilities 454
5 Likelihood ratio tests 455
6 Example 456
References 458
Using Observed Functional Data to Simulate a Stochastic Process via a Random Multiplicative Cascade Model 459
1 Introduction 459
2 The Random Multiplicative Cascade Model 460
3 The Adjustment Curve for Binary Response of a Simulated Stochastic Process 462
4 Application 463
5 Conclusion and perspectives 465
Acknowledgements. 466
References 466
A Clusterwise Center and Range Regression Model for Interval-Valued Data 467
1 Introduction 467
2 A Brief Overview of Regression for Interval-Valued Data and Clusterwise Linear Regression 468
2.1 Regression for Interval Data 468
2.2 Clusterwise Linear Regression 468
3 Clusterwise regression on interval-valued data 469
3.1 Step 1: definition of the best prototypes 470
3.2 Step 2: definition of the best partition 471
4 Application: a car interval-valued data set 472
5 Concluding Remarks 474
References 474
Contributions to Bayesian Structural Equation Modeling 475
1 Structural equation models 475
1.1 Context 475
1.2 Model 475
1.3 The role of latent variables and identifiability constraints 476
2 Bayesian estimation of SEM 476
2.1 Bayesian estimation 476
2.2 Conditional posterior distribution of latent variables 476
2.3 Conditional posterior distributions of parameters 477
2.4 The Gibbs sampler 478
2.5 Validation 479
2.6 Identifiability issues 479
3 Application 480
4 Conclusion and perspectives 481
Acknowledgements 482
References 482
Some Examples of Statistical Computing in France During the 19th Century 483
1 Introduction 483
2 Covariance analysis 484
2.1 Constant intercept and different slopes 484
2.2 Different intercept and constant slopes 485
3 Cauchy’s heuristic for regression 485
3.1 Cauchy’s linear estimators for simple regression 485
3.2 Cauchy’s multiple regression 486
3.3 Spurious correlation and differencing 486
4 Iteratively weighted least squares 487
5 Unsuccessful attempts 487
5.1 A weighted mean with data driven weights 488
5.2 Fitting a Gamma density 488
6 Concluding remarks 489
References 490
Imputation by Gaussian Copula Model with an Application to Incomplete Customer Satisfaction Data 491
1 Introduction 491
2 Preliminaries 492
3 Imputation algorithms based on Gaussian copula 493
4 Imputing incomplete customer satisfaction data 495
5 Concluding remarks 497
Appendix 1. Proof of Lemma 1 497
Acknowledgments 498
References 498
On Multiple-Case Diagnostics in Linear Subspace Method 499
1 Introduction 499
2 Sensitivity analysis in linear subspace method 499
2.1 CLAFIC 500
2.2 Discriminant score 500
2.3 EIF and SIF 501
2.4 Diagnostics 501
3 A multiple-case diagnostics with clustering 502
4 Numerical examples 503
5 Concluding Remarks 506
References 506
Fourier Methods for Sequential Change Point Analysis in Autoregressive Models 507
1 Introduction 507
2 Test statistics 508
3 Sketch of asymptotics 509
4 FLS estimation and ECF statistics 510
5 Computational issues 512
Acknowlegements: 513
References 513
Computational Treatment of the Error Distribution in Nonparametric Regression with Right-Censored and Selection-Biased Data 515
1 Introduction and model 515
2 Description of the method 517
3 Practical implementation and simulations 518
3.1 Bandwidth selection procedure 518
3.2 Simulations 519
4 Data analysis 521
Acknowledgements. 522
References 522
Mixtures of Weighted Distance-Based Models for Ranking Data 523
1 Introduction 523
2 Distance-based models for ranking data 524
2.1 Distance-based models 524
3 Mixtures of weighted distance-based models 525
3.1 Weighted distance-based models 525
3.2 Mixture models 526
4 Simulation Studies 527
5 Application to real data: social science research on political goals 528
6 Conclusion 530
Acknowledgement 530
References 530
Fourier Analysis and Swarm Intelligence for Stochastic Optimization of Discrete Functions 531
1 Introduction 531
2 Computational method 532
2.1 Discrete fourier analysis 532
2.2 Particle Swarm Optimization 533
3 Results and discussion 534
4 Conclusions 536
References 538
Global Hypothesis Test to Simultaneously Compare the Predictive Values of Two Binary Diagnostic Tests in Paired Designs: a Simul 539
1 Introduction 539
2 Global hypothesis test 540
3 Simulation experiments 542
4 Conclusions 545
References 546
Modeling Operational Risk: Estimation and Effects of Dependencies 547
1 Introduction 547
2 Correlation 548
3 Tail dependence 549
3.1 Estimation via copulas 549
3.2 Nonparametric estimation 550
4 Effects on risk–capital estimation 552
5 Conclusion 553
References 554
Learning Hierarchical Bayesian Networks for Genome-Wide Association Studies 555
1 Introduction 555
2 Background for HLC model learning 556
3 Constructing the FHLC model 557
3.1 Principle 557
3.2 Node partitioning 558
3.3 Parameter learning and missing data imputation 558
3.4 Controlling information decay 559
4 Sketch of algorithm CFHLC 559
5 Experimental results and discussion 560
6 Concluding remarks 562
References 562
Exact Posterior Distributions over the Segmentation Space and Model Selection for Multiple Change-Point Detection Problems 563
1 Introduction 563
2 Exploring the segmentation space 565
2.2 Posterior distribution of the change-points and segments 566
2.3 Posterior entropy 567
3 Model selection 567
3.1 Exact BIC criterion 567
3.2 ICL criterion for dimension selection 568
4 Applications 568
4.1 Simulation study 568
4.2 Analysis of a CGH profile 569
References 569
Parcellation Schemes and Statistical Tests to Detect Active Regions on the Cortical Surface 571
1 Introduction 571
2 Model 572
2.1 Inputs and notations 572
2.2 Hierarchical parcellations 573
2.3 Estimation of the model 574
2.4 Optimizing the parameters of the model 575
2.5 Random-effects (RFX) inference procedure 575
3 Results on a real dataset 575
4 Discussion 577
References 578
Robust Principal Component Analysis Based on Pairwise Correlation Estimators 579
1 Introduction 579
2 Pairwise correlation estimators 580
2.1 Univariate trimming. 580
2.2 Bivariate trimming. 581
2.3 Gnanadesikan-Kettenring estimator. 581
2.4 Quadrant correlation. 581
3 Simulation Study 582
Bivariate correlation outliers: 582
Multivariate correlation outliers: 582
Componentwise outliers: 582
4 Conclusion 584
References 585
Ordinary Least Squares for Histogram Data Based on Wasserstein Distance 587
1 Introduction 587
Assumption of error-free representation of macro-units 588
First two moments are finite 588
Internal independence 588
Histogram variable 588
Empirical distribution function variable 588
Model of distribution variable 588
2 Histogram Data and Wasserstein Distance 588
3 The Model 590
3.1 Tools for the evaluation of the Goodness of Fit of the model 591
4 Application 593
5 Conclusions 593
References 594
DetMCD in a Calibration Framework 595
1 Introduction 595
2 The DetMCD algorithm 596
2.1 General procedure 596
2.2 Initial estimates 597
3 Robust calibration methods 598
4 Simulation study 598
4.1 Simulation design 599
4.2 Results 600
5 Summary and conclusion 602
References 602
Separable Two-Dimensional Linear Discriminant Analysis 603
1 Introduction 603
2 Review of LDA and 2DLDA 605
2.1 Review of LDA 605
2.2 Review of 2DLDA 605
3 Separable 2DLDA 606
3.1 Motivation: Using separable covariance matrix for 2D data 606
3.3 Maximum likelihood estimation of separable covariance matrix 607
4 Experiments 608
5 Conclusion 609
Acknowledgement 609
References 610
List of Supplementary Contributed and Invited Papers Only Available on springerlink.com 611
Index 623