Hierarchical conditional random fields for parts based models matching

A parts based model is a parametrization of an object class using a collection of landmarks following the object structure. The matching of parts based models is one of the problems where pairwise Conditional Random Fields have been successfully applied. The main reason of their effectiveness is tractable inference and learning due to the simplicity of involved graphs, usually trees. However, these models do not consider possible patterns of statistics among sets of landmarks, and thus they sufffer from using too myopic information. To overcome this limitation, we propoese a novel structure based on a hierarchical Conditional Random Fields, which we explain in the first part of this memory. We build a hierarchy of combinations of landmarks, where matching is performed taking into account the whole hierarchy. To preserve tractable inference we effectively sample the label set. We test our method on facial feature selection and human pose estimation on two challenging datasets: Buffy and MultiPIE. In the second part of this memory, we present a novel approach to multiple kernel combination that relies on stacked classification. This method can be used to evaluate the landmarks of the parts-based model approach. Our method is based on combining responses of a set of independent classifiers for each individual kernel. Unlike earlier approaches that linearly combine kernel responses, our approach uses them as inputs to another set of classifiers. We will show that we outperform state-of-the-art methods on most of the standard benchmark datasets.

Detecting the number of clusters of individuals using the software STRUCTURE: a simulation study.

The identification of genetically homogeneous groups of individuals is a long standing issue in population genetics. A recent Bayesian algorithm implemented in the software STRUCTURE allows the identification of such groups. However, the ability of this algorithm to detect the true number of clusters (K) in a sample of individuals when patterns of dispersal among populations are not homogeneous has not been tested. The goal of this study is to carry out such tests, using various dispersal scenarios from data generated with an individual-based model. We found that in most cases the estimated 'log probability of data' does not provide a correct estimation of the number of clusters, K. However, using an ad hoc statistic DeltaK based on the rate of change in the log probability of data between successive K values, we found that STRUCTURE accurately detects the uppermost hierarchical level of structure for the scenarios we tested. As might be expected, the results are sensitive to the type of genetic marker used (AFLP vs. microsatellite), the number of loci scored, the number of populations sampled, and the number of individuals typed in each sample.

Policy Diffusion: An Agent-Based Approach

Methods like Event History Analysis can show the existence of diffusion and part of its nature, but do not study the process itself. Nowadays, thanks to the increasing performance of computers, processes can be studied using computational modeling. This thesis presents an agent-based model of policy diffusion mainly inspired from the model developed by Braun and Gilardi (2006). I first start by developing a theoretical framework of policy diffusion that presents the main internal drivers of policy diffusion - such as the preference for the policy, the effectiveness of the policy, the institutional constraints, and the ideology - and its main mechanisms, namely learning, competition, emulation, and coercion. Therefore diffusion, expressed by these interdependencies, is a complex process that needs to be studied with computational agent-based modeling. In a second step, computational agent-based modeling is defined along with its most significant concepts: complexity and emergence. Using computational agent-based modeling implies the development of an algorithm and its programming. When this latter has been developed, we let the different agents interact. Consequently, a phenomenon of diffusion, derived from learning, emerges, meaning that the choice made by an agent is conditional to that made by its neighbors. As a result, learning follows an inverted S-curve, which leads to partial convergence - global divergence and local convergence - that triggers the emergence of political clusters; i.e. the creation of regions with the same policy. Furthermore, the average effectiveness in this computational world tends to follow a J-shaped curve, meaning that not only time is needed for a policy to deploy its effects, but that it also takes time for a country to find the best-suited policy. To conclude, diffusion is an emergent phenomenon from complex interactions and its outcomes as ensued from my model are in line with the theoretical expectations and the empirical evidence.Les méthodes d'analyse de biographie (event history analysis) permettent de mettre en évidence l'existence de phénomènes de diffusion et de les décrire, mais ne permettent pas d'en étudier le processus. Les simulations informatiques, grâce aux performances croissantes des ordinateurs, rendent possible l'étude des processus en tant que tels. Cette thèse, basée sur le modèle théorique développé par Braun et Gilardi (2006), présente une simulation centrée sur les agents des phénomènes de diffusion des politiques. Le point de départ de ce travail met en lumière, au niveau théorique, les principaux facteurs de changement internes à un pays : la préférence pour une politique donnée, l'efficacité de cette dernière, les contraintes institutionnelles, l'idéologie, et les principaux mécanismes de diffusion que sont l'apprentissage, la compétition, l'émulation et la coercition. La diffusion, définie par l'interdépendance des différents acteurs, est un système complexe dont l'étude est rendue possible par les simulations centrées sur les agents. Au niveau méthodologique, nous présenterons également les principaux concepts sous-jacents aux simulations, notamment la complexité et l'émergence. De plus, l'utilisation de simulations informatiques implique le développement d'un algorithme et sa programmation. Cette dernière réalisée, les agents peuvent interagir, avec comme résultat l'émergence d'un phénomène de diffusion, dérivé de l'apprentissage, où le choix d'un agent dépend en grande partie de ceux faits par ses voisins. De plus, ce phénomène suit une courbe en S caractéristique, poussant à la création de régions politiquement identiques, mais divergentes au niveau globale. Enfin, l'efficacité moyenne, dans ce monde simulé, suit une courbe en J, ce qui signifie qu'il faut du temps, non seulement pour que la politique montre ses effets, mais également pour qu'un pays introduise la politique la plus efficace. En conclusion, la diffusion est un phénomène émergent résultant d'interactions complexes dont les résultats du processus tel que développé dans ce modèle correspondent tant aux attentes théoriques qu'aux résultats pratiques.

Trabecular Bone Score: A Noninvasive Analytical Method Based Upon the DXA Image.

The trabecular bone score (TBS) is a gray-level textural metric that can be extracted from the two-dimensional lumbar spine dual-energy X-ray absorptiometry (DXA) image. TBS is related to bone microarchitecture and provides skeletal information that is not captured from the standard bone mineral density (BMD) measurement. Based on experimental variograms of the projected DXA image, TBS has the potential to discern differences between DXA scans that show similar BMD measurements. An elevated TBS value correlates with better skeletal microstructure; a low TBS value correlates with weaker skeletal microstructure. Lumbar spine TBS has been evaluated in cross-sectional and longitudinal studies. The following conclusions are based upon publications reviewed in this article: 1) TBS gives lower values in postmenopausal women and in men with previous fragility fractures than their nonfractured counterparts; 2) TBS is complementary to data available by lumbar spine DXA measurements; 3) TBS results are lower in women who have sustained a fragility fracture but in whom DXA does not indicate osteoporosis or even osteopenia; 4) TBS predicts fracture risk as well as lumbar spine BMD measurements in postmenopausal women; 5) efficacious therapies for osteoporosis differ in the extent to which they influence the TBS; 6) TBS is associated with fracture risk in individuals with conditions related to reduced bone mass or bone quality. Based on these data, lumbar spine TBS holds promise as an emerging technology that could well become a valuable clinical tool in the diagnosis of osteoporosis and in fracture risk assessment. © 2014 American Society for Bone and Mineral Research.

A model of cellular dosimetry for macroscopic tumors in radiopharmaceutical therapy.

PURPOSE: In the radiopharmaceutical therapy approach to the fight against cancer, in particular when it comes to translating laboratory results to the clinical setting, modeling has served as an invaluable tool for guidance and for understanding the processes operating at the cellular level and how these relate to macroscopic observables. Tumor control probability (TCP) is the dosimetric end point quantity of choice which relates to experimental and clinical data: it requires knowledge of individual cellular absorbed doses since it depends on the assessment of the treatment's ability to kill each and every cell. Macroscopic tumors, seen in both clinical and experimental studies, contain too many cells to be modeled individually in Monte Carlo simulation; yet, in particular for low ratios of decays to cells, a cell-based model that does not smooth away statistical considerations associated with low activity is a necessity. The authors present here an adaptation of the simple sphere-based model from which cellular level dosimetry for macroscopic tumors and their end point quantities, such as TCP, may be extrapolated more reliably. METHODS: Ten homogenous spheres representing tumors of different sizes were constructed in GEANT4. The radionuclide 131I was randomly allowed to decay for each model size and for seven different ratios of number of decays to number of cells, N(r): 1000, 500, 200, 100, 50, 20, and 10 decays per cell. The deposited energy was collected in radial bins and divided by the bin mass to obtain the average bin absorbed dose. To simulate a cellular model, the number of cells present in each bin was calculated and an absorbed dose attributed to each cell equal to the bin average absorbed dose with a randomly determined adjustment based on a Gaussian probability distribution with a width equal to the statistical uncertainty consistent with the ratio of decays to cells, i.e., equal to Nr-1/2. From dose volume histograms the surviving fraction of cells, equivalent uniform dose (EUD), and TCP for the different scenarios were calculated. Comparably sized spherical models containing individual spherical cells (15 microm diameter) in hexagonal lattices were constructed, and Monte Carlo simulations were executed for all the same previous scenarios. The dosimetric quantities were calculated and compared to the adjusted simple sphere model results. The model was then applied to the Bortezomib-induced enzyme-targeted radiotherapy (BETR) strategy of targeting Epstein-Barr virus (EBV)-expressing cancers. RESULTS: The TCP values were comparable to within 2% between the adjusted simple sphere and full cellular models. Additionally, models were generated for a nonuniform distribution of activity, and results were compared between the adjusted spherical and cellular models with similar comparability. The TCP values from the experimental macroscopic tumor results were consistent with the experimental observations for BETR-treated 1 g EBV-expressing lymphoma tumors in mice. CONCLUSIONS: The adjusted spherical model presented here provides more accurate TCP values than simple spheres, on par with full cellular Monte Carlo simulations while maintaining the simplicity of the simple sphere model. This model provides a basis for complementing and understanding laboratory and clinical results pertaining to radiopharmaceutical therapy.

Trabecular bone score: a noninvasive analytical method based upon the DXA image.

The trabecular bone score (TBS) is a gray-level textural metric that can be extracted from the two-dimensional lumbar spine dual-energy X-ray absorptiometry (DXA) image. TBS is related to bone microarchitecture and provides skeletal information that is not captured from the standard bone mineral density (BMD) measurement. Based on experimental variograms of the projected DXA image, TBS has the potential to discern differences between DXA scans that show similar BMD measurements. An elevated TBS value correlates with better skeletal microstructure; a low TBS value correlates with weaker skeletal microstructure. Lumbar spine TBS has been evaluated in cross-sectional and longitudinal studies. The following conclusions are based upon publications reviewed in this article: 1) TBS gives lower values in postmenopausal women and in men with previous fragility fractures than their nonfractured counterparts; 2) TBS is complementary to data available by lumbar spine DXA measurements; 3) TBS results are lower in women who have sustained a fragility fracture but in whom DXA does not indicate osteoporosis or even osteopenia; 4) TBS predicts fracture risk as well as lumbar spine BMD measurements in postmenopausal women; 5) efficacious therapies for osteoporosis differ in the extent to which they influence the TBS; 6) TBS is associated with fracture risk in individuals with conditions related to reduced bone mass or bone quality. Based on these data, lumbar spine TBS holds promise as an emerging technology that could well become a valuable clinical tool in the diagnosis of osteoporosis and in fracture risk assessment.

Labor markets and monetary policy: A new-Keynesian model with unemployement

We construct a utility-based model of fluctuations, with nominal rigidities andunemployment, and draw its implications for the unemployment-inflation trade-off and for the conduct of monetary policy.We proceed in two steps. We first leave nominal rigidities aside. We show that,under a standard utility specification, productivity shocks have no effect onunemployment in the constrained efficient allocation. We then focus on theimplications of alternative real wage setting mechanisms for fluctuations in un-employment. We show the role of labor market frictions and real wage rigiditiesin determining the effects of productivity shocks on unemployment.We then introduce nominal rigidities in the form of staggered price setting byfirms. We derive the relation between inflation and unemployment and discusshow it is influenced by the presence of labor market frictions and real wagerigidities. We show the nature of the tradeoff between inflation and unemployment stabilization, and its dependence on labor market characteristics. We draw the implications for optimal monetary policy.

Further insights on the French WISC-IV factor structure through Bayesian structural equation modeling

The interpretation of the Wechsler Intelligence Scale for Children-Fourth Edition (WISC-IV) is based on a 4-factor model, which is only partially compatible with the mainstream Cattell-Horn-Carroll (CHC) model of intelligence measurement. The structure of cognitive batteries is frequently analyzed via exploratory factor analysis and/or confirmatory factor analysis. With classical confirmatory factor analysis, almost all crossloadings between latent variables and measures are fixed to zero in order to allow the model to be identified. However, inappropriate zero cross-loadings can contribute to poor model fit, distorted factors, and biased factor correlations; most important, they do not necessarily faithfully reflect theory. To deal with these methodological and theoretical limitations, we used a new statistical approach, Bayesian structural equation modeling (BSEM), among a sample of 249 French-speaking Swiss children (8-12 years). With BSEM, zero-fixed cross-loadings between latent variables and measures are replaced by approximate zeros, based on informative, small-variance priors. Results indicated that a direct hierarchical CHC-based model with 5 factors plus a general intelligence factor better represented the structure of the WISC-IV than did the 4-factor structure and the higher order models. Because a direct hierarchical CHC model was more adequate, it was concluded that the general factor should be considered as a breadth rather than a superordinate factor. Because it was possible for us to estimate the influence of each of the latent variables on the 15 subtest scores, BSEM allowed improvement of the understanding of the structure of intelligence tests and the clinical interpretation of the subtest scores.

Hybrid consumption paths in the attribute space: A model and application with scanner data

This paper presents and estimates a dynamic choice model in the attribute space considering rational consumers. In light of the evidence of several state-dependence patterns, the standard attribute-based model is extended by considering a general utility function where pure inertia and pure variety-seeking behaviors can be explained in the model as particular linear cases. The dynamics of the model are fully characterized by standard dynamic programming techniques. The model presents a stationary consumption pattern that can be inertial, where the consumer only buys one product, or a variety-seeking one, where the consumer shifts among varied products.We run some simulations to analyze the consumption paths out of the steady state. Underthe hybrid utility assumption, the consumer behaves inertially among the unfamiliar brandsfor several periods, eventually switching to a variety-seeking behavior when the stationary levels are approached. An empirical analysis is run using scanner databases for three different product categories: fabric softener, saltine cracker, and catsup. Non-linear specifications provide the best fit of the data, as hybrid functional forms are found in all the product categories for most attributes and segments. These results reveal the statistical superiority of the non-linear structure and confirm the gradual trend to seek variety as the level of familiarity with the purchased items increases.

On The Spot Damage Detection Methodology for Highway Bridges, July 2010

Vibration-based damage identification (VBDI) techniques have been developed in part to address the problems associated with an aging civil infrastructure. To assess the potential of VBDI as it applies to highway bridges in Iowa, three applications of VBDI techniques were considered in this study: numerical simulation, laboratory structures, and field structures. VBDI techniques were found to be highly capable of locating and quantifying damage in numerical simulations. These same techniques were found to be accurate in locating various types of damage in a laboratory setting with actual structures. Although there is the potential for these techniques to quantify damage in a laboratory setting, the ability of the methods to quantify low-level damage in the laboratory is not robust. When applying these techniques to an actual bridge, it was found that some traditional applications of VBDI methods are capable of describing the global behavior of the structure but are most likely not suited for the identification of typical damage scenarios found in civil infrastructure. Measurement noise, boundary conditions, complications due to substructures and multiple material types, and transducer sensitivity make it very difficult for present VBDI techniques to identify, much less quantify, highly localized damage (such as small cracks and minor changes in thickness). However, while investigating VBDI techniques in the field, it was found that if the frequency-domain response of the structure can be generated from operating traffic load, the structural response can be animated and used to develop a holistic view of the bridge’s response to various automobile loadings. By animating the response of a field bridge, concrete cracking (in the abutment and deck) was correlated with structural motion and problem frequencies (i.e., those that cause significant torsion or tension-compression at beam ends) were identified. Furthermore, a frequency-domain study of operational traffic was used to identify both common and extreme frequencies for a given structure and loading. Common traffic frequencies can be compared to problem frequencies so that cost-effective, preventative solutions (either structural or usage-based) can be developed for a wide range of IDOT bridges. Further work should (1) perfect the process of collecting high-quality operational frequency response data; (2) expand and simplify the process of correlating frequency response animations with damage; and (3) develop efficient, economical, preemptive solutions to common damage types.

Measurement Errors Should Always Be Incorporated in Phylogenetic Comparative Analysis

The evolution of continuous traits is the central component of comparative analyses in phylogenetics, and the comparison of alternative models of trait evolution has greatly improved our understanding of the mechanisms driving phenotypic differentiation. Several factors influence the comparison of models, and we explore the effects of random errors in trait measurement on the accuracy of model selection. We simulate trait data under a Brownian motion model (BM) and introduce different magnitudes of random measurement error. We then evaluate the resulting statistical support for this model against two alternative models: Ornstein-Uhlenbeck (OU) and accelerating/decelerating rates (ACDC). Our analyses show that even small measurement errors (10%) consistently bias model selection towards erroneous rejection of BM in favour of more parameter-rich models (most frequently the OU model). Fortunately, methods that explicitly incorporate measurement errors in phylogenetic analyses considerably improve the accuracy of model selection. Our results call for caution in interpreting the results of model selection in comparative analyses, especially when complex models garner only modest additional support. Importantly, as measurement errors occur in most trait data sets, we suggest that estimation of measurement errors should always be performed during comparative analysis to reduce chances of misidentification of evolutionary processes.

APPRAISAL OF THE SNAP MODEL FOR PREDICTING NITROGEN MINERALIZATION IN TROPICAL SOILS UNDER EUCALYPTUS

The Soil Nitrogen Availability Predictor (SNAP) model predicts daily and annual rates of net N mineralization (NNM) based on daily weather measurements, daily predictions of soil water and soil temperature, and on temperature and moisture modifiers obtained during aerobic incubation (basal rate). The model was based on in situ measurements of NNM in Australian soils under temperate climate. The purpose of this study was to assess this model for use in tropical soils under eucalyptus plantations in São Paulo State, Brazil. Based on field incubations for one month in three, NNM rates were measured at 11 sites (0-20 cm layer) for 21 months. The basal rate was determined in in situ incubations during moist and warm periods (January to March). Annual rates of 150-350 kg ha-1 yr-1 NNM predicted by the SNAP model were reasonably accurate (R2 = 0.84). In other periods, at lower moisture and temperature, NNM rates were overestimated. Therefore, if used carefully, the model can provide adequate predictions of annual NNM and may be useful in practical applications. For NNM predictions for shorter periods than a year or under suboptimal incubation conditions, the temperature and moisture modifiers need to be recalibrated for tropical conditions.

Benchmarking therapeutic drug monitoring software: A systematic evaluation of available computer tools

Introduction: Therapeutic drug monitoring (TDM) aims at optimizing treatment by individualizing dosage regimen based on measurement of blood concentrations. Maintaining concentrations within a target range requires pharmacokinetic and clinical capabilities. Bayesian calculation represents a gold standard in TDM approach but requires computing assistance. In the last decades computer programs have been developed to assist clinicians in this assignment. The aim of this benchmarking was to assess and compare computer tools designed to support TDM clinical activities.¦Method: Literature and Internet search was performed to identify software. All programs were tested on common personal computer. Each program was scored against a standardized grid covering pharmacokinetic relevance, user-friendliness, computing aspects, interfacing, and storage. A weighting factor was applied to each criterion of the grid to consider its relative importance. To assess the robustness of the software, six representative clinical vignettes were also processed through all of them.¦Results: 12 software tools were identified, tested and ranked. It represents a comprehensive review of the available software's characteristics. Numbers of drugs handled vary widely and 8 programs offer the ability to the user to add its own drug model. 10 computer programs are able to compute Bayesian dosage adaptation based on a blood concentration (a posteriori adjustment) while 9 are also able to suggest a priori dosage regimen (prior to any blood concentration measurement), based on individual patient covariates, such as age, gender, weight. Among those applying Bayesian analysis, one uses the non-parametric approach. The top 2 software emerging from this benchmark are MwPharm and TCIWorks. Other programs evaluated have also a good potential but are less sophisticated (e.g. in terms of storage or report generation) or less user-friendly.¦Conclusion: Whereas 2 integrated programs are at the top of the ranked listed, such complex tools would possibly not fit all institutions, and each software tool must be regarded with respect to individual needs of hospitals or clinicians. Interest in computing tool to support therapeutic monitoring is still growing. Although developers put efforts into it the last years, there is still room for improvement, especially in terms of institutional information system interfacing, user-friendliness, capacity of data storage and report generation.

A model of cellular dosimetry for macroscopic tumors in radiopharmaceutical therapy.

PURPOSE: In the radiopharmaceutical therapy approach to the fight against cancer, in particular when it comes to translating laboratory results to the clinical setting, modeling has served as an invaluable tool for guidance and for understanding the processes operating at the cellular level and how these relate to macroscopic observables. Tumor control probability (TCP) is the dosimetric end point quantity of choice which relates to experimental and clinical data: it requires knowledge of individual cellular absorbed doses since it depends on the assessment of the treatment's ability to kill each and every cell. Macroscopic tumors, seen in both clinical and experimental studies, contain too many cells to be modeled individually in Monte Carlo simulation; yet, in particular for low ratios of decays to cells, a cell-based model that does not smooth away statistical considerations associated with low activity is a necessity. The authors present here an adaptation of the simple sphere-based model from which cellular level dosimetry for macroscopic tumors and their end point quantities, such as TCP, may be extrapolated more reliably. METHODS: Ten homogenous spheres representing tumors of different sizes were constructed in GEANT4. The radionuclide 131I was randomly allowed to decay for each model size and for seven different ratios of number of decays to number of cells, N(r): 1000, 500, 200, 100, 50, 20, and 10 decays per cell. The deposited energy was collected in radial bins and divided by the bin mass to obtain the average bin absorbed dose. To simulate a cellular model, the number of cells present in each bin was calculated and an absorbed dose attributed to each cell equal to the bin average absorbed dose with a randomly determined adjustment based on a Gaussian probability distribution with a width equal to the statistical uncertainty consistent with the ratio of decays to cells, i.e., equal to Nr-1/2. From dose volume histograms the surviving fraction of cells, equivalent uniform dose (EUD), and TCP for the different scenarios were calculated. Comparably sized spherical models containing individual spherical cells (15 microm diameter) in hexagonal lattices were constructed, and Monte Carlo simulations were executed for all the same previous scenarios. The dosimetric quantities were calculated and compared to the adjusted simple sphere model results. The model was then applied to the Bortezomib-induced enzyme-targeted radiotherapy (BETR) strategy of targeting Epstein-Barr virus (EBV)-expressing cancers. RESULTS: The TCP values were comparable to within 2% between the adjusted simple sphere and full cellular models. Additionally, models were generated for a nonuniform distribution of activity, and results were compared between the adjusted spherical and cellular models with similar comparability. The TCP values from the experimental macroscopic tumor results were consistent with the experimental observations for BETR-treated 1 g EBV-expressing lymphoma tumors in mice. CONCLUSIONS: The adjusted spherical model presented here provides more accurate TCP values than simple spheres, on par with full cellular Monte Carlo simulations while maintaining the simplicity of the simple sphere model. This model provides a basis for complementing and understanding laboratory and clinical results pertaining to radiopharmaceutical therapy.

An Experimental Validation of a Statistical-Based Damage Detection Approach, January 2011

In this work, a previously-developed, statistical-based, damage-detection approach was validated for its ability to autonomously detect damage in bridges. The damage-detection approach uses statistical differences in the actual and predicted behavior of the bridge caused under a subset of ambient trucks. The predicted behavior is derived from a statistics-based model trained with field data from the undamaged bridge (not a finite element model). The differences between actual and predicted responses, called residuals, are then used to construct control charts, which compare undamaged and damaged structure data. Validation of the damage-detection approach was achieved by using sacrificial specimens that were mounted to the bridge and exposed to ambient traffic loads and which simulated actual damage-sensitive locations. Different damage types and levels were introduced to the sacrificial specimens to study the sensitivity and applicability. The damage-detection algorithm was able to identify damage, but it also had a high false-positive rate. An evaluation of the sub-components of the damage-detection methodology and methods was completed for the purpose of improving the approach. Several of the underlying assumptions within the algorithm were being violated, which was the source of the false-positives. Furthermore, the lack of an automatic evaluation process was thought to potentially be an impediment to widespread use. Recommendations for the improvement of the methodology were developed and preliminarily evaluated. These recommendations are believed to improve the efficacy of the damage-detection approach.