2063 research outputs found
Sort by
Competing risk modelling for in-hospital length of stay
Get PDFIn this study, we propose a framework for analysing in-hospital patient data from electronic health records. We transform longitudinal sparse vital signs measurements into cross-sectional data via descriptive statistics, imputing missing values, and evaluating variables strongly associated with time to mutually exclusive events (favourable medical discharge or deterioration). We employ competing risk and random survival forest techniques to predict patients’ length of stay and evaluate models’ performance via Brier score
Predicting the arrival of the unpredictable: An approach for foreseeing the transition to chaos of wildfire propagation
Get PDFA discrete map for modelling wildfire propagation is derived from a prototypical reaction-diffusion equation for the temperature field. We show that, for a constant fuel concentration at the fire-front, the heat transfer coefficient from fuel to surroundings and as well as an effective heat of reaction are two independent mechanisms that can cause the transition to chaos, when they may depend on temperature as a consequence of the fire-atmosphere coupling and of the fuel inhomogeneity, respectively. In particular, chaos can enter when the coefficient for the heat transfer from the fuel to the surrounding depends linearly on the temperature and when the effective heat of reaction depends quadratically. Moreover, when the fuel concentration field at the fire-front fluctuates, this embodies a third mechanism that may cause the transition to chaos even without any fire-atmosphere coupling or fuel inhomogeneity. Surprisingly, when the effective heat of reaction depends linearly on the temperature, the chaos generated by the non-constant fuel concentration is ceased. This suppression is not observed when the chaos is due to the fire-atmosphere coupling with constant fuel concentration. In all cases, the onset of chaos is related to the logistic map. The application of this approach for setting an alternative method for real-time risk assessment is discussed in the conclusions.This research is supported by the Basque Government through the BERC 2022–2025 program, by the Ministry of Science and Innovation: BCAM Severo Ochoa accreditation CEX2021-001142-S / MICIN / AEI / 10.13039/501100011033 and the project PID2019-107685RB-I00, and by the Spanish State Research Agency (AEI) through the project PDC2022-133115-I00 entitled ”B 2 F 2: Be a Better digital Fire-Fighter” and funded by the European Union Next Generation EU; by the European Regional Development Fund (ERDF) and the Department of Education of the regional government, the Junta of Castilla y Le ́on (Grant contract SA089P20)
An exponential integration generalized multiscale finite element method for parabolic problems
Get PDFWe consider linear and semilinear parabolic problems posed in high-contrast multiscale media in two dimensions.
The presence of high-contrast multiscale media adversely affects the accuracy, stability, and overall efficiency of
numerical approximations such as finite elements in space combined with some time integrator. In many cases,
implementing time discretizations such as finite differences or exponential integrators may be impractical because
each time iteration needs the computation of matrix operators involving very large and ill-conditioned sparse matrices.
Here, we propose an efficient Generalized Multiscale Finite Element Method (GMsFEM) that is robust against the
high-contrast diffusion coefficient. We combine GMsFEM with exponential integration in time to obtain a good
approximation of the final time solution. Our approach is efficient and practical because it computes matrix functions
of small matrices given by the GMsFEM method. We present representative numerical experiments that show the
advantages of combining exponential integration and GMsFEM approximations. The constructions and methods
developed here can be easily adapted to three-dimensional domains
Spatio‑temporal modelling of high‑throughput phenotyping data
Get PDFHigh throughput phenotyping (HTP) platforms and devices are increasingly used to characterise growth and developmental processes for large sets of plant genotypes. This dissertation is motivated by the need to accurately estimate genetic effects over time when analysing data from such HTP experiments. The HTP data we deal with here are characterised by phenotypic traits measured multiple times in the presence of spatial and temporal noise and a hierarchical organisation at three levels (populations, genotypes within populations, and plants within genotypes). The challenge is to balance efficient statistical models and com- putational solutions to deal with the complexity and dimensionality of the experimental data. To that aim, we propose two strategies. The first proposal divides the problem into two stages. The first stage (spatial model) focuses on correcting the phenotypic data for experimental design factors and spatial variation, while the second stage (hierarchical longitudinal model) aims to estimate the evolution over time of the genetic signal. The second proposal is to face the problem simultaneously (one-stage approach). That is, mod- elling the longitudinal evolution of the genetic effect on a given phenotypic trait while accounting for the temporal and spatial effects of environmental and design factors (spatio-temporal hierarchical model). We follow the same modelling philosophy throughout our work and propose multidimensional P-spline-based hierarchical approaches. We provide the user with appealing tools that take advantage of the sparse model matrices structure to reduce computational complexity. All our codes are publicly available on the R-package statgenHTP and https://gitlab.bcamath.org/dperez/htp_one_stage_approach. We illustrate the performance of our methods using spatio-temporal simulated data and data from the PhenoArch greenhouse platform at INRAE Montpellier and the outdoor Field Phenotyping platform at ETH Zürich. In the plant breeding context, we show how to extract new time-independent phenotypes for genomic selection purposes.MTM2017-82379-R
BERC 2018-2021
BERC 2022-2025
SEV-2017-0718
CEX2021-001142-S/MICIN/AEI/10.13039/50110001103
Mapping flagellated swimmers to surface-slip driven swimmers.
Get PDFFlagellated microswimmers are ubiquitous in natural habitats. Understanding the hydrodynamic behavior of these cells is of paramount interest, owing to their applications in
bio-medical engineering and disease spreading. Since the last two decades, computational
efforts have been continuously improved to accurately capture the complex hydrodynamic
behavior of these model systems. However, modeling the dynamics of such swimmers with
fine details is computationally expensive due to the large number of unknowns and the small
time-steps required to solve the equations. In this work we propose a method to map fully
resolved flagellated microswimmers to coarse, active slip driven swimmers which can be simulated at a reduced computational cost. Using the new method, the slip driven swimmers
move with the same velocity, to machine precision, as the flagellated swimmers and generate a similar flow field with a controlled accuracy. The method is validated for swimming
patterns near a no-slip boundary, interactions between swimmers and scattering with large
obstacles
Learning a logistic regression with the help of unknown features at prediction stage
Get PDFThe use of features available at training time, but not
at prediction time, as additional information for training models
is known as learning using privileged information paradigm. In
this paper, the handling of privileged features is addressed from
the logistic regression perspective, commonly used in the clinical
setting. Two new proposals, LOGIT+ and LRPROB+, learned
with the influence of privileged features and preserving the
interpretability of conventional logistic regression, are proposed.
Experimental results on datasets report improvements of our
proposals over the performance of traditional logistic regression
learned without privileged information
Fast parallel IGA-ADS solver for time-dependent Maxwell's equations
Get PDFWe propose a simulator for time-dependent Maxwell's equations with linear computational cost. We employ B-spline basis functions as considered in the isogeometric analysis (IGA). We focus on non-stationary Maxwell's equations defined on a regular patch of elements. We employ the idea of alternating-directions splitting (ADS) and employ a second-order accurate time-integration scheme for the time-dependent Maxwell's equations in a weak form. After discretization, the resulting stiffness matrix exhibits a Kronecker product structure. Thus, it enables linear computational cost LU factorization. Additionally, we derive a formulation for absorbing boundary conditions (ABCs) suitable for direction splitting. We perform numerical simulations of the scattering problem (traveling pulse wave) to verify the ABC. We simulate the radiation of electromagnetic (EM) waves from the dipole antenna. We verify the order of the time integration scheme using a manufactured solution problem. We then simulate magnetotelluric measurements. Our simulator is implemented in a shared memory parallel machine, with the GALOIS library supporting the parallelization. We illustrate the parallel efficiency with strong and weak scalability tests corresponding to non-stationary Maxwell simulations.EXPERTIA (KK-2021/00048)
SIGZE (KK-2021/00095)
PDC2021-121093-I0
APPLICATION OF SPARSE DICTIONARY LEARNING TO SEISMIC DATA RECONSTRUCTION
Get PDFAccording to the principle of compressed sensing (CS), under-sampled seismic data can be interpolated when the data becomes sparse in a transform domain. To sparsify the data, dictionary learning presents a data-driven approach trained to be optimized for each target dataset. This study presents an interpolation method for seismic data in which dictionary learning is employed to improve the sparsity of data representation using improved Kth Singular Value Decomposition (K-SVD). In this way, the transformation will be highly compatible with the input data, and the data in the converted domain will be sparser. In addition, the sampling matrix is produced with the restricted isometry property (RIP). To reduce the sensitivity of the minimizer term to the outliers, we use the smooth L1 minimizer as a regularization term in the regularized orthogonal matching pursuit (ROMP). We apply the proposed method to both synthetic and real seismic data. The results show that it can successfully reconstruct the missing seismic traces
A semi-supervised algorithm for improving the consistency of crowdsourced datasets: The COVID-19 case study on respiratory disorder classification
Get PDFBackground and Objective: Cough audio signal classification is a potentially useful tool in screening for respiratory disorders, such as COVID-19. Since it is dangerous to collect data from patients with contagious diseases, many research teams have turned to crowdsourcing to quickly gather cough sound data. The COUGHVID dataset enlisted expert physicians to annotate and diagnose the underlying diseases present in a limited number of recordings. However, this approach suffers from potential cough mislabeling, as well as disagreement between experts. Methods: In this work, we use a semi-supervised learning (SSL) approach – based on audio signal processing tools and interpretable machine learning models – to improve the labeling consistency of the COUGHVID dataset for 1) COVID-19 versus healthy cough sound classification 2) distinguishing wet from dry coughs, and 3) assessing cough severity. First, we leverage SSL expert knowledge aggregation techniques to overcome the labeling inconsistencies and label sparsity in the dataset. Next, our SSL approach is used to identify a subsample of re-labeled COUGHVID audio samples that can be used to train or augment future cough classifiers. Results: The consistency of the re-labeled COVID-19 and healthy data is demonstrated in that it exhibits a high degree of inter-class feature separability: 3x higher than that of the user-labeled data. Similarly, the SSL method increases this separability by 11.3x for cough type and 5.1x for severity classifications. Furthermore, the spectral differences in the user-labeled audio segments are amplified in the re-labeled data, resulting in significantly different power spectral densities between healthy and COVID-19 coughs in the 1-1.5 kHz range (p=1.2×10−64), which demonstrates both the increased consistency of the new dataset and its explainability from an acoustic perspective. Finally, we demonstrate how the re-labeled dataset can be used to train a COVID-19 classifier, achieving an AUC of 0.797. Conclusions: We propose a SSL expert knowledge aggregation technique for the field of cough sound classification for the first time, and demonstrate how it can be used to combine the medical knowledge of multiple experts in an explainable fashion, thus providing abundant, consistent data for cough classification tasks.RYC2021-032853-
Weak-type maximal function estimates on the infinite-dimensional torus
Get PDFWe prove necessary and sufficient conditions for the weak- boundedness, for , of a maximal operator on the infinite-dimensional torus. In the endpoint case we obtain the same weak-type inequality enjoyed by the strong maximal function in dimension two. Our results are quantitatively sharp.Ikerbasque
RYC2018-025477-I
START 032.202