Learning from Geometry

Subspaces and manifolds are two powerful models for high dimensional signals. Subspaces model linear correlation and are a good fit to signals generated by physical systems, such as frontal images of human faces and multiple sources impinging at an antenna array. Manifolds model sources that are not linearly correlated, but where signals are determined by a small number of parameters. Examples are images of human faces under different poses or expressions, and handwritten digits with varying styles. However, there will always be some degree of model mismatch between the subspace or manifold model and the true statistics of the source. This dissertation exploits subspace and manifold models as prior information in various signal processing and machine learning tasks.

A near-low-rank Gaussian mixture model measures proximity to a union of linear or affine subspaces. This simple model can effectively capture the signal distribution when each class is near a subspace. This dissertation studies how the pairwise geometry between these subspaces affects classification performance. When model mismatch is vanishingly small, the probability of misclassification is determined by the product of the sines of the principal angles between subspaces. When the model mismatch is more significant, the probability of misclassification is determined by the sum of the squares of the sines of the principal angles. Reliability of classification is derived in terms of the distribution of signal energy across principal vectors. Larger principal angles lead to smaller classification error, motivating a linear transform that optimizes principal angles. This linear transformation, termed TRAIT, also preserves some specific features in each class, being complementary to a recently developed Low Rank Transform (LRT). Moreover, when the model mismatch is more significant, TRAIT shows superior performance compared to LRT.

The manifold model enforces a constraint on the freedom of data variation. Learning features that are robust to data variation is very important, especially when the size of the training set is small. A learning machine with large numbers of parameters, e.g., deep neural network, can well describe a very complicated data distribution. However, it is also more likely to be sensitive to small perturbations of the data, and to suffer from suffer from degraded performance when generalizing to unseen (test) data.

From the perspective of complexity of function classes, such a learning machine has a huge capacity (complexity), which tends to overfit. The manifold model provides us with a way of regularizing the learning machine, so as to reduce the generalization error, therefore mitigate overfiting. Two different overfiting-preventing approaches are proposed, one from the perspective of data variation, the other from capacity/complexity control. In the first approach, the learning machine is encouraged to make decisions that vary smoothly for data points in local neighborhoods on the manifold. In the second approach, a graph adjacency matrix is derived for the manifold, and the learned features are encouraged to be aligned with the principal components of this adjacency matrix. Experimental results on benchmark datasets are demonstrated, showing an obvious advantage of the proposed approaches when the training set is small.

Stochastic optimization makes it possible to track a slowly varying subspace underlying streaming data. By approximating local neighborhoods using affine subspaces, a slowly varying manifold can be efficiently tracked as well, even with corrupted and noisy data. The more the local neighborhoods, the better the approximation, but the higher the computational complexity. A multiscale approximation scheme is proposed, where the local approximating subspaces are organized in a tree structure. Splitting and merging of the tree nodes then allows efficient control of the number of neighbourhoods. Deviation (of each datum) from the learned model is estimated, yielding a series of statistics for anomaly detection. This framework extends the classical {\em changepoint detection} technique, which only works for one dimensional signals. Simulations and experiments highlight the robustness and efficacy of the proposed approach in detecting an abrupt change in an otherwise slowly varying low-dimensional manifold.

Performance and Robustness Analysis of Co-Prime and Nested Sampling

Coprime and nested sampling are well known deterministic sampling techniques that operate at rates significantly lower than the Nyquist rate, and yet allow perfect reconstruction of the spectra of wide sense stationary signals. However, theoretical guarantees for these samplers assume ideal conditions such as synchronous sampling, and ability to perfectly compute statistical expectations. This thesis studies the performance of coprime and nested samplers in spatial and temporal domains, when these assumptions are violated. In spatial domain, the robustness of these samplers is studied by considering arrays with perturbed sensor locations (with unknown perturbations). Simplified expressions for the Fisher Information matrix for perturbed coprime and nested arrays are derived, which explicitly highlight the role of co-array. It is shown that even in presence of perturbations, it is possible to resolve $O(M^2)$ under appropriate conditions on the size of the grid. The assumption of small perturbations leads to a novel ``bi-affine" model in terms of source powers and perturbations. The redundancies in the co-array are then exploited to eliminate the nuisance perturbation variable, and reduce the bi-affine problem to a linear underdetermined (sparse) problem in source powers. This thesis also studies the robustness of coprime sampling to finite number of samples and sampling jitter, by analyzing their effects on the quality of the estimated autocorrelation sequence. A variety of bounds on the error introduced by such non ideal sampling schemes are computed by considering a statistical model for the perturbation. They indicate that coprime sampling leads to stable estimation of the autocorrelation sequence, in presence of small perturbations. Under appropriate assumptions on the distribution of WSS signals, sharp bounds on the estimation error are established which indicate that the error decays exponentially with the number of samples. The theoretical claims are supported by extensive numerical experiments.

Random perturbations of stochastic processes with unbounded variable length memory

We consider binary infinite order stochastic chains perturbed by a random noise. This means that at each time step, the value assumed by the chain can be randomly and independently flipped with a small fixed probability. We show that the transition probabilities of the perturbed chain are uniformly close to the corresponding transition probabilities of the original chain. As a consequence, in the case of stochastic chains with unbounded but otherwise finite variable length memory, we show that it is possible to recover the context tree of the original chain, using a suitable version of the algorithm Context, provided that the noise is small enough.

Resurgence of inner solutions for perturbations of the McMillan map

A sequence of “inner equations” attached to certain perturbations of the McMillan map was considered in [MSS09], their solutions were used in that article to measure an exponentially small separatrix splitting. We prove here all the results relative to these equations which are necessary to complete the proof of the main result of [MSS09]. The present work relies on ideas from resurgence theory: we describe the formal solutions, study the analyticity of their Borel transforms and use ´Ecalle’s alien derivations to measure the discrepancy between different Borel-Laplace sums.

Nonperturbative semiclassical stability of de Sitter spacetime for small metric deviations

We consider the linearized semiclassical Einstein equations for small deviations around de Sitter spacetime including the vacuum polarization effects of conformal fields. Employing the method of order reduction, we find the exact solutions for general metric perturbations (of scalar, vector and tensor type). Our exact (nonperturbative) solutions show clearly that in this case de Sitter is stable with respect to small metric deviations and a late-time attractor. Furthermore, they also reveal a breakdown of perturbative solutions for a sufficiently long evolution inside the horizon. Our results are valid for any conformal theory, even self-interacting ones with arbitrarily strong coupling.

Evaluation of steroidomics by liquid chromatography hyphenated to mass spectrometry as a powerful analytical strategy for measuring human steroid perturbations.

This review presents the evolution of steroid analytical techniques, including gas chromatography coupled to mass spectrometry (GC-MS), immunoassay (IA) and targeted liquid chromatography coupled to mass spectrometry (LC-MS), and it evaluates the potential of extended steroid profiles by a metabolomics-based approach, namely steroidomics. Steroids regulate essential biological functions including growth and reproduction, and perturbations of the steroid homeostasis can generate serious physiological issues; therefore, specific and sensitive methods have been developed to measure steroid concentrations. GC-MS measuring several steroids simultaneously was considered the first historical standard method for analysis. Steroids were then quantified by immunoassay, allowing a higher throughput; however, major drawbacks included the measurement of a single compound instead of a panel and cross-reactivity reactions. Targeted LC-MS methods with selected reaction monitoring (SRM) were then introduced for quantifying a small steroid subset without the problems of cross-reactivity. The next step was the integration of metabolomic approaches in the context of steroid analyses. As metabolomics tends to identify and quantify all the metabolites (i.e., the metabolome) in a specific system, appropriate strategies were proposed for discovering new biomarkers. Steroidomics, defined as the untargeted analysis of the steroid content in a sample, was implemented in several fields, including doping analysis, clinical studies, in vivo or in vitro toxicology assays, and more. This review discusses the current analytical methods for assessing steroid changes and compares them to steroidomics. Steroids, their pathways, their implications in diseases and the biological matrices in which they are analysed will first be described. Then, the different analytical strategies will be presented with a focus on their ability to obtain relevant information on the steroid pattern. The future technical requirements for improving steroid analysis will also be presented.

Turbulent shallow-water model for orographic subgrid-scale perturbations

A parallel pseudo-spectral method for the simulation in distributed memory computers of the shallow-water equations in primitive form was developed and used on the study of turbulent shallow-waters LES models for orographic subgrid-scale perturbations. The main characteristics of the code are: momentum equations integrated in time using an accurate pseudo-spectral technique; Eulerian treatment of advective terms; and parallelization of the code based on a domain decomposition technique. The parallel pseudo-spectral code is efficient on various architectures. It gives high performance onvector computers and good speedup on distributed memory systems. The code is being used for the study of the interaction mechanisms in shallow-water ows with regular as well as random orography with a prescribed spectrum of elevations. Simulations show the evolution of small scale vortical motions from the interaction of the large scale flow and the small-scale orographic perturbations. These interactions transfer energy from the large-scale motions to the small (usually unresolved) scales. The possibility of including the parametrization of this effects in turbulent LES subgrid-stress models for the shallow-water equations is addressed.

Dynamique des communautés végétales et impacts des perturbations humaines sur la végétation des tourbières.

Ce mémoire visait à comprendre la dynamique temporelle et les patrons floristiques actuels de deux tourbières du sud-ouest du Québec (Small et Large Tea Field) et à identifier les facteurs anthropiques, environnementaux et spatiaux sous-jacents. Pour répondre aux objectifs, des inventaires floristiques anciens (1985) ont d’abord été comparés à des inventaires récents (2012) puis les patrons actuels et les facteurs sous-jacents ont été identifiés à l’aide d’analyses multi-variables. Mes résultats montrent d’abord qu’un boisement important s’est produit au cours des 30 dernières années dans les tourbières à l’étude, probablement en lien avec le drainage des terres agricoles avoisinantes, diminuant la hauteur de la nappe phréatique. Simultanément, les sphaignes ont proliférées dans le centre des sites s’expliquant par une recolonisation des secteurs ayant brûlés en 1983. D’autre part, mes analyses ont montré que les patrons floristiques actuels étaient surtout liés aux variables environnementales (pH et conductivité de l’eau, épaisseur des dépôts), bien que la variance associée aux activités humaines était aussi significative, notamment dans la tourbière Large (18.6%). Les patrons floristiques ainsi que les variables environnementales et anthropiques explicatives étaient aussi fortement structurés dans l’espace, notamment selon un gradient bordure-centre. Enfin, la diversité béta actuelle était surtout liée à la présence d’espèces non-tourbicoles ou exotiques. Globalement, cette étude a montré que les perturbations humaines passées et actuelles avaient un impact important sur la dynamique et la distribution de la végétation des tourbières Small et Large Tea Field.

The spacing of orographic rainbands triggered by small-scale topography

A combination of idealized numerical simulations and analytical theory is used to investigate the spacing between convective orographic rainbands over the Coastal Range of western Oregon. The simulations, which are idealized from an observed banded precipitation event over the Coastal Range, indicate that the atmospheric response to conditionally unstable flow over the mountain ridge depends strongly on the subridge-scale topographic forcing on the windward side of the ridge. When this small-scale terrain contains only a single scale (l) of terrain variability, the band spacing is identical to l, but when a spectrum of terrain scales are simultaneously present, the band spacing ranges between 5 and 10 km, a value that is consistent with observations. Based on the simulations, an inviscid linear model is developed to provide a physical basis for understanding the scale selection of the rainbands. This analytical model, which captures the transition from lee waves upstream of the orographic cloud to moist convection within it, reveals that the spacing of orographic rainbands depends on both the projection of lee-wave energy onto the unstable cap cloud and the growth rate of unstable perturbations within the cloud. The linear model is used in tandem with numerical simulations to determine the sensitivity of the band spacing to a number of environmental and terrain-related parameters.

The impact of salinity perturbations on the future uptake of heat by the Atlantic Ocean

Anthropogenic ocean heat uptake is a key factor in determining climate change and sea-level rise. There is considerable uncertainty in projections of freshwater forcing of the ocean, with the potential to influence ocean heat uptake. We investigatethis by adding either -0.1 Sv or +0.1 Sv freshwater to the Atlantic in global climate model simulations, simultaneously imposing an atmospheric CO2 increase. The resulting changes in the Atlantic meridional overturning circulation are roughly equal and opposite (±2Sv). The impact of the perturbation on ocean heat content is more complex, although it is relatively small (~5%) compared to the total anthropogenic heat uptake. Several competing processes either accelerate or retard warming at different depths. Whilst positive freshwater perturbations cause an overall heating of the Atlantic, negative perturbations produce insignificant net changes in heat content. The processes active in our model appear robust, although their net result is likely model- and experiment-dependent.

Forest-obligate Sabethes mosquitoes suggest palaeoecological perturbations

The origin of tropical forest diversity has been hotly debated for decades. Although specific mechanisms vary, many such explanations propose some vicariance in the distribution of species during glacial cycles and several have been supported by genetic evidence in Neotropical taxa. However, no consensus exists with regard to the extent or time frame of the vicariance events. Here, we analyse the cytochrome oxidase II mitochondrial gene of 250 Sabethes albiprivus B mosquitoes sampled from western Sao Paulo in Brazil. There was very low population structuring among collection sites (Phi(ST) = 0.03, P = 0.04). Historic demographic analyses and the contemporary geographic distribution of genetic diversity suggest that the populations sampled are not at demographic equilibrium. Three distinct mitochondrial clades were observed in the samples, one of which differed significantly in its geographic distribution relative to the other two within a small sampling area (similar to 70 x 35 km). This fact, supported by the inability of maximum likelihood analyses to achieve adequate fits to simple models for the population demography of the species, suggests a more complex history, possibly involving disjunct forest refugia. This hypothesis is supported by a genetic signal of recent population growth, which is expected if population sizes of this forest-obligate insect increased during the forest expansions that followed glacial periods. Although a time frame cannot be reliably inferred for the vicariance event leading to the three genetic clades, molecular clock estimates place this at similar to 1 Myr before present.

Suppression of small baryonic structures due to a primordial magnetic field

We investigate the impact of the existence of a primordial magnetic field on the filter mass, characterizing the minimum baryonic mass that can form in dark matter (DM) haloes. For masses below the filter mass, the baryon content of DM haloes are severely depressed. The filter mass is the mass when the baryon to DM mass ratio in a halo is equal to half the baryon to DM ratio of the Universe. The filter mass has previously been used in semi-analytic calculations of galaxy formation, without taking into account the possible existence of a primordial magnetic field. We examine here its effect on the filter mass. For homogeneous comoving primordial magnetic fields of B(0) similar to 1 or 2 nG and a re-ionization epoch that starts at a redshift z(s) = 11 and is completed at z(r) = 8, the filter mass is increased at redshift 8, for example, by factors of 4.1 and 19.8, respectively. The dependence of the filter mass on the parameters describing the re-ionization epoch is investigated. Our results are particularly important for the formation of low-mass galaxies in the presence of a homogeneous primordial magnetic field. For example, for B(0) similar to 1 nG and a re-ionization epoch of z(s) similar to 11 and z(r) similar to 7, our results indicate that galaxies of total mass M similar to 5 x 108 M(circle dot) need to form at redshifts z(F) greater than or similar to 2.0, and galaxies of total mass M similar to 108 M(circle dot) at redshifts z(F) greater than or similar to 7.7.

The signature of dark energy perturbations in galaxy cluster surveys

Models of dynamical dark energy unavoidably possess fluctuations in the energy density and pressure of that new component. In this paper we estimate the impact of dark energy fluctuations on the number of galaxy clusters in the Universe using a generalization of the spherical collapse model and the Press-Schechter formalism. The observations we consider are several hypothetical Sunyaev-Zel`dovich and weak lensing (shear maps) cluster surveys, with limiting masses similar to ongoing (SPT, DES) as well as future (LSST, Euclid) surveys. Our statistical analysis is performed in a 7-dimensional cosmological parameter space using the Fisher matrix method. We find that, in some scenarios, the impact of these fluctuations is large enough that their effect could already be detected by existing instruments such as the South Pole Telescope, when priors from other standard cosmological probes are included. We also show how dark energy fluctuations can be a nuisance for constraining cosmological parameters with cluster counts, and point to a degeneracy between the parameter that describes dark energy pressure on small scales (the effective sound speed) and the parameters describing its equation of state.

The signature of dark energy perturbations in galaxy cluster surveys

Fundação de Amparo à Pesquisa do Estado de São Paulo (FAPESP)

The IFUSP race-track microtron booster accelerator end magnets antisymmetric perturbations

We will present measurements and calculations related to the antisymmetric perturbations, and comparisons with the symmetric ones, of the IFUSP race-track microtron booster accelerator end magnets. These perturbations were measured in planes situated at +/-12 mm of the middle plane, in a gap height of 4 cm, for a field distribution of about 0.1 T. The measurements were done in 1170 points, separated by a distance of 8 mm, using an automated system with a +/-1.5 mu T differential Hall probe. The race-track microtron booster is the second stage of the 30.0 MeV electron accelerator under construction at the Linear Accelerator Laboratory in which the required uniformity for the magnetic field is of about 10(-3). The method of correction employed to homogenize the IFUSP race-track microtron booster accelerator magnets assures uniformity of 10(-5) in an average field of 0.1 T, over an area of 700 cm(2). This method uses the principle of attaching to the pole pieces correction coils produced by etching techniques, with copper leads shaped like the isofield lines of the normal component of the magnetic field measured. The ideal planes, in which these measurements are done, are calculated and depend on the behavior of the magnetic field perturbations: symmetric or antisymmetric with reference to the middle plane of the magnet gap. These calculations are presented in this work and show that for antisymmetric perturbations there is no ideal plane for the correction of the magnetic field; for the symmetric one, these planes are at +/-60% of the half gap height, from the middle plane. So this method of correction is not feasible for antisymmetric perturbations, as will be shown. Besides, the correction of the symmetric portion of the field distribution does not influence the antisymmetric one, which almost does not change, and corroborates the theoretical predictions. We found antisymmetric perturbations of small intensity only in one of the two end magnets. However, they are not detected at +/- 1 mm of the middle plane and will not damage the electron beam.