2063 research outputs found
Sort by
Advances in learning using privileged information for supervised classification
Full text linkThe world is increasingly data-dependent with more automatic jobs thanks to advances in the development of intelligent machines capable of learning complex patterns accurately. However, in some scenarios, not all available information is leveraged in the machine learning process. In a clinical setting, a patient’s health record is available, but typically not all of the information is used during the learning process. For instance, some features may be discarded from the training process due to their unavailability at deployment time. These features are called privileged features and are leveraged through the Learning Using Privileged Information (LUPI) paradigm, a learning scenario that exploits privileged features as additional information for training models. In this dissertation, we establish new methodologies that incorporate the LUPI paradigm with supervised classification algorithms.
First, motivated by the widespread use of logistic regression in the clinical field, we introduce two new methods that leverage privileged information for learning a logistic regression. For its development, the parameters of a conventional logistic regression trained with all available features, privileged and regular, are projected onto the parameter space associated to regular features (available at training and deployment time). The projection to obtain the model parameters is performed by the minimization of two different loss functions governed by logit terms and posterior probabilities. In addition, a metric is proposed to determine whether the use of privileged information can enhance performance.
Subsequently, the privileged learning process is addressed through a knowledge distillation perspective: information from a teacher learned with regular and privileged features is transferred to a student composed exclusively of regular features. While most approaches assume perfect knowledge for the teacher, it can commit mistakes. Assuming that, we propose a novel privileged distillation framework with a double contribution. Firstly, a designed function to imitate the teacher when it classifies correctly and to differ in cases of misclassification. Secondly, an adaptation of the cross-entropy loss to appropriately penalize the instances where the student outperforms the teacher. Its effectiveness is empirically demonstrated on datasets with imperfect teachers, significantly enhancing the performance of state-of-the-art frameworks. Furthermore, necessary conditions for successful privileged learning are presented, along with a dataset categorization based on the information provided by the privileged features.
Finally, we propose a multi-task privileged framework that combines two types of tasks. First, the privileged-prediction task involves using regular features to predict privileged information, working as an intermediate step to guide the learning process. Second, the main learning objective, the target task, uses the predicted privileged information along with the regular features to make the final target prediction. Furthermore, knowledge distillation techniques are included within the target task to effectively exploit the privileged information. Additionally, we analyze misclassification causes and refine the proposed multi-task privileged learning to reduce errors.
All the contributions reported in the dissertation are empirically evaluated in tabular datasets and image-related problems, achieving improvements compared to state-of-the-art approaches.PID2022-137442NB-I00
PRE2021-09927
Novel P-Spline-based approaches for curve and derivative estimation in longitudinal growth studies
Full text linkLongitudinal data analysis plays a key role in understanding individual development over time. It helps identify important characteristics such as growth patterns, biological variations, and key milestones like growth spurts and the age at which they occur. Although several statistical models and software implementations are available to estimate growth curves and their derivatives, challenges remain due to the complex and dynamic nature of these processes.
This dissertation explores a range of statistical methods for modeling growth based on repeated measurements from individuals. In particular, it highlights the use of penalized spline (P-spline) models, which have gained popularity for their flexibility in modeling smooth curves. Additionally, their connection with mixed-effects models allows us to take advantage of the methodological and computational tools in this setting for fitting P-spline models.
We focus on three main classes of models for fitting individual growth trajectories and estimating their derivatives: (i) the double-penalty P-spline model, (ii) a family of semiparametric mixed-effects models, and (iii) modified versions of the SuperImposition by Translation and Rotation (SITAR) model. Each approach is discussed in terms of its theoretical background, practical implementation, and suitability for growth analysis.
The proposed methods are evaluated through simulation studies and compared with alternative approaches to assess their performance. Our results indicate that the proposed models provide a slight improvement in curve fitting and derivative estimation. Additionally, we demonstrate the application of these methods using real longitudinal data from 125 young professional football players to analyze their growth and maturation processes over time.PID2020-115882RB-I0
Mesoscale modelling of fibrin clots: The interplay between rheology and microstructure at the gel point
Full text linkThis study presents a numerical model for incipient fibrin-clot formation that captures characteristic rheological and microstructural features of the clot at the gel point. Using a mesoscale-clustering framework, we evaluate the effect of gel concentration or gel volume fraction and branching on the fractal dimension, the gel time, and the viscoelastic properties of the clots. We show that variations in the gel concentration of our model can reproduce the effect of thrombin in the formation of fibrin clots. In particular, the model reproduces the fractal dimension's dependency on gel concentration and the trends in elasticity and gelation time with varying thrombin concentrations. This approach allows us to accurately recreate the gelation point of fibrin-thrombin gels, highlighting the intricate process of fibrin polymerization and gel network formation. This is critical for applications in the clinical and bioengineering fields where precise control over the gelation process is required
Morphological Transitions of Block Copolymer Micelles: Implications for Mesoporous Materials Ordering
Full text linkThe design of block-copolymer-based functional materials, including mesoporous membranes and nanoparticles, requires a comprehensive understanding of the hierarchical assembly of block copolymers in selective solvents into micelles and subsequent ordered phases. It is hypothesized that micellar ordering and characteristic assembly can be described using a set of phase parameters that account for entropic and enthalpic interactions. Dissipative particle dynamics (DPD) simulations are used to systematically investigate the self-assembly of semidiluted block copolymers, resembling isoporous membrane preparation conditions. The effect of Flory–Huggins interaction parameters, block lengths, and concentration on the morphology and polydispersity of the micelles is evaluated. The interaction parameters are mapped into Flory–Huggins theory by considering the block's conformation. These results reveal the effect of polymer concentration and solvent affinity on the morphological transition of the aggregates, in agreement with existing experimental evidence. It is identified that monodisperse-spherical micelles in solution are fundamental to stabilize ordered states. Weak solvent segregation of the largest block, curvature of the core-corona interface, and stretching of the corona-forming one are found to be key to stabilize monodisperse assemblies. These conditions can be predicted using spherical-micelles packing considerations and a global phase parameter from the Flory–Huggins theory. This study provides valuable insights into the self-assembly of diblock copolymers and offers a potential way to optimize the preparation of mesoporous ordered structures and micelle ordering in semidiluted systems
Multi-scale modeling of Snail-mediated response to hypoxia in tumor progression
No full textTumor cell migration within the microenvironment is a crucial aspect for cancer progression and, in this context, hypoxia has a significant role. An inadequate oxygen supply acts as an environmental stressor inducing migratory bias and phenotypic changes. In this paper, we propose a novel multi-scale mathematical model to analyze the pivotal role of Snail protein expression in the cellular responses to hypoxia. Starting from the description of single-cell dynamics driven by the Snail protein, we construct the corresponding kinetic transport equation that describes the evolution of the cell distribution. Subsequently, we employ proper scaling arguments to formally derive the equations for the statistical moments of the cell distribution, which govern the macroscopic tumor dynamics. Numerical simulations of the model are performed in various scenarios with biological relevance to provide insights into the role of the multiple tactic terms, the impact of Snail expression on cell proliferation, and the emergence of hypoxia-induced migration patterns. Moreover, quantitative comparisons with experimental data show the model's reliability in measuring the impact of Snail transcription on cell migratory potential. Through our findings, we shed light on the potential of our mathematical framework in advancing the understanding of the biological mechanisms driving tumor progression.Italian Ministry of Education, Universities and Research, MIUR grant Dipartimento di Eccellenza 2018-2022, project E11G18000350001 (MC, GC, MD)
National Group of Mathematical Physics (GNFM-INdAM), INdAM–GNFM Project (CUP E53C22001930001) "From kinetic to macroscopic models for tumor-immune system competition"
Modeling Nature Research Unit, Grant QUAL21-011. Consejería de Universidad, Investigaciòn e Innovaciòn (Junta de Andalucía). City of Hope’s Global Scholar Progra
Entanglement transfer during quantum frequency conversion in gas-filled hollow-core fibers
No full textQuantum transduction is essential for the future hybrid quantum networks, connecting devices across different spectral ranges. In this regard, molecular modulation in hollow-core fibers has proven to be exceptional for efficient and tunable frequency conversion of arbitrary light fields down to the single-photon limit. However, insights on this conversion method for quantum light have remained elusive beyond standard semiclassical models. In this Letter, we employ a quantum Hamiltonian framework to characterize the behavior of entanglement during molecular modulation, while describing the quantum dynamics of both molecules and photons in agreement with recent experiments. In particular, apart from obtaining analytical expressions for the final opto-molecular states, our model predicts a close correlation between the evolution of the average photon numbers and the transfer of entanglement between the interacting parties. These results will contribute to the development of new fiber-based strategies to tackle the challenges associated with the upcoming generation of lightwave quantum technologies.We acknowledge financial support from HORIZON-CL4-2022-QUANTUM01-SGA project 101113946 OpenSuperQ-Plus100 of the EU Flagship on Quantum Technologies, the Spanish Ramón y Cajal Grant RYC-2020-030503-I, project Grants No. PID2021-125823NA-I00, PID2021-123131NA-I00, PID2021-122505OBC31 and TED2021-129959B-C21, funded by MICIU/AEI/10.13039/501100011033 and by “ERDF A way of making Europe”, by “ERDF Invest in your Future”, by the “European Union NextGenerationEU/PRTR” and "ESF+", from the Basque Government through Grants No. IT1470-22 and IT1455-22 and ELKARTEK (4Smart-KK-2023/00016, Ekohegaz II-KK-2023/00051, and KUBIT KK-2024/00105), and from the IKUR Strategy under the collaboration agreement between Ikerbasque Foundation and BCAM on behalf of the Department of Education of the Basque Government and the grant IKUR\_IKA\_23/04. ML acknowledges support from the predoctoral grant "Formación de Profesorado Universitario" FPU23/02350 from the Spanish Ministry of Science, Innovation and Universities (MICIU). This work has also been financially supported by the Ministry for Digital Transformation and the Civil Service of the Spanish Government through the QUANTUM ENIA project call – Quantum Spain project, and by the European Union through the Recovery, Transformation and Resilience Plan – NextGenerationEU within the framework of the Digital Spain 2026 Agenda
Quantifying the impact of COVID-19 non-pharmaceutical interventions and vaccination using mathematical modeling and cost-effectiveness analysis
Full text linkIn 2020, COVID-19 emerged as a global health crisis, prompting the widespread adoption of non-pharmaceutical interventions (NPIs) such as lockdowns, social distancing, and mask mandates. These measures, along with large-scale vaccination efforts, aimed to control virus transmission and reduce severe health outcomes among the elderly and those with comorbidities, albeit with considerable economic impacts. This dissertation examines the effectiveness and cost-effectiveness of public health measures against COVID-19, with a focus on the Basque Country—a region with a highly aging population. Through mathematical modeling and health economics, we assess the role of NPIs in reducing transmission, the impact of the disease in life expectancy, as well as the cost-effectiveness of vaccination, particularly given the spread of more transmissible variants. This work presents estimates of hospitalizations and deaths prevented by vaccination, using models fitted to two years of daily data, and compares these with findings from other settings. These insights inform a cost-effectiveness analysis from a healthcare payer perspective
Efficient quantum amplitude encoding of polynomial functions
Full text linkLoading functions into quantum computers represents an essential step in several quantum algorithms, such as quantum partial differential equation solvers. Therefore, the inefficiency of this process leads to a major bottleneck for the application of these algorithms. Here, we present and compare two efficient methods for the amplitude encoding of real polynomial functions on n qubits. This case holds special relevance, as any continuous function on a closed interval can be uniformly approximated with arbitrary precision by a polynomial function. The first approach relies on the matrix product state representation (MPS). We study and benchmark the approximations of the target state when the bond dimension is assumed to be small. The second algorithm combines two subroutines. Initially we encode the linear function into the quantum registers either via its MPS or with a shallow sequence of multi-controlled gates that loads the linear function’s Hadamard-Walsh series, and we explore how truncating the Hadamard-Walsh series of the linear function affects the final fidelity. Applying the inverse discrete Hadamard-Walsh transform converts the state encoding the series coefficients into an amplitude encoding of the linear function. Thus, we use this construction as a building block to achieve an exact block encoding of the amplitudes corresponding to the linear function on k0 qubits and apply the quantum singular value transformation that implements a polynomial transformation to the block encoding of the amplitudes. This unitary together with the Amplitude Amplification algorithm will enable us to prepare the quantum state that encodes the polynomial function on k0 qubits. Finally we pad n − k0 qubits to generate an approximated encoding of the polynomial on n qubits, analyzing the error depending on k0. In this regard, our methodology proposes a method to improve the state-of-the-art complexity by introducing controllable errors.The authors acknowledge financial support from OpenSuperQ+100 (Grant No. 101113946) of the EU Flagship on Quantum Technologies, as well as from the EU FET-Open project EPIQUS (Grant No. 899368), also from Project Grant No. PID2021-125823NA-I00 595 and Spanish Ramón y Cajal Grant No. RYC-2020-030503-I funded by MCIN/AEI/10.13039/501100011033 and by “ERDF A way of making Europe” and “ERDF Invest in your Future,” this project has also received support from the Spanish Ministry for Digital Transformation and of Civil Service of the Spanish Government through the QUANTUM ENIA project call - Quantum Spain, and by the EU through the Recovery, Transformation and Resilience Plan – NextGenerationEU within the framework of the Digital Spain 2026 Agenda, we acknowledge funding from Basque Government through Grant No. IT1470-22 and the IKUR Strategy under the collaboration agreement between Ikerbasque Foundation and BCAM on behalf of the Department of Education of the Basque Government, as well as from and UPV/EHU Ph.D. Grant No. PIF20/276. PR acknowledges financial support from the CDTI within the Misiones 2021 program and the Ministry of Science and Innovation under the Recovery, Transformation and Resilience Plan—Next Generation EU under the project “CUCO: Quantum Computing and its Application to Strategic Industries
