Research Projects

Ongoing Research Projects

IRIS: Hacia un asistente personal para la mejora del bienestar emocional
Ayudas para la realización de doctorados industriales 2023; Comunidad de Madrid; 2024-2026

 

The objective of the project is to develop the necessary algorithms for the implementation of a personalized assistant (IRIS) for improving the mental health and emotional well-being based on artificial intelligence (AI). To achieve this, research will be conducted to develop a prototype (or minimum viable product) that includes all the elements of IRIS and allows for commercial use while providing data for its complete development.


IRIS is based on automatic intervention cycles in 10 predefined areas of intervention, derived from information collected from mobile devices. These 10 areas define the physical and emotional health status of a user and include sleep, social skills, emotional state, personal organization, mobile phone usage, physical activity, nutrition, stress and work, cognition, and existing pathologies (previously diagnosed conditions). The intervention cycle in each of these areas consists of a range of suggestions or resources that vary based on the indicators obtained from each area through the use of mobile phones, applications, and other wearables. These range from messages of positive reinforcement if the indicators are excellent, to recommending seeking a specialist if the indicators suggest it..

Prevención de reCAídas en tRastornos dE conducta. – PreCARE
Agencia Estatal de Investigación (AEI); 2023 – 2026

 

The main objective of PreCARE is to develop a solution for prevention and intervention against mental health relapses and their improvement in care by combining:

a) Development of methods for profiling patients, detecting behavioral changes and explaining these changes for these pathologies.

b) Implementation of these methods in the eB2 MindCare platform and their application to health care

c) Development and validation of the solution in four units (addictions, dual pathology, rehabilitation and return to the community and eating disorders) in two hospitals

Almacenaje y Transmisión de Información sobre Redes Autónomas y Distribuídas (ADICT-Net)
Agencia Estatal de Investigación (AEI); 2023 – 2026

 

El proyecto combina técnicas de áreas como Teoría de la Información, codificación y redes de comunicaciones inalámbricas para mejorar la velocidad y latencia de las redes modernas, como por ejemplo el internet de las cosas, redes vehiculares y Fog Computing. Para conseguir esto, perseguiremos dos vías en paralelo: analítica y computacional. La vía analítica buscará modelos matemáticos y estadísticos para los canales y solicitudes de los usuarios. Después, técnicas de optimización se usarán para diseñar la mejor transmisión en cada situación. La vía computacional utilizará una mezcla de aprendizaje máquina profundo y poco profundo para llevar a cabo la misma tarea. Además, planeamos implementar los protocolos en una red de Software Defined Radios (SDRs) y probarlos en simulaciones reales a través del aire. Específicamente, las principales contribuciones de este proyecto son:

1) Extender la teoría del almacenamiento codificado a escenarios con contenido distribuído, flujos interferentes y canales variables,

2) Diseñar protocolos de almacenaje y transmisión nuevos y más eficientes para redes inalámbricas,

3) Analizar el potencial de aplicar técnicas de aprendizaje máquina a resolver varias cuestiones en redes de comunicaciones,

4) Implementar una librería en código abierto para evaluar protocolos de redes utilizando SDRs.

Medicina Personalizada (MedPer) en la detección precoz del deterioro cognitivo (DC) preclinico. Desarrollo de un modelo predictivo de riesgo.
Instituto de Salud Carlos III PERTE Salud (IBSAL); 2023 – 2025

 

Generación de modelos predictivos para la detección precoz de deterioro cognitivo que permitan su seguimiento para prevenir (en lo posible) o retrasar su aparición. Consorcio formado por investigadores con la experiencia y objetivo común de implementar la MedPer para la detección del deterioro cognitivo en un estadio preclínico (antes de la aparición de sintomatología clínica) y con utilidad en la práctica clínica.

Integrating logitudinal patient-generated data and multi-omic profiling for comprehensive precision oncology in Womens’ cancers
Instituto de Salud Carlos III PERTE Salud (CNIO); 2023 – 2025

 

A comprehensive precision oncology approach that integrates personalized genomics and individualized PDU collection will allow an unprecedented level of understanding of cancer processes, tackling the features that drive patient disease trajectories and outcomes, eliciting truly precision interventions. We term this new wave of precision oncology Patient-Led Precision Oncology (PLPO), and we expect PLPO will help achieve the 2 overarching goals: improve our current predictive ability, and break current efficacy plateaus. Our specific objectives are:

-To capture and integrate the PDU in a cohort of patients with women’s cancers.
-To establish patient disease trajectories and identify features that forecast individual outcomes.
-To gain biological knowledge resolution (molecular taxonomy) of seemingly identical outcomes across patients.
-To develop Patient Digital Twin (PDT) models enabling testing interventions that pinpoint individual actionable features that improve
outcomes, as a potential tool for mid-term clinical implementation in the advanced cancer patient clinical decision tree.

This integrative project combines translational and clinical oncology, engineering, data science and novel artificial intelligence (AI) approaches, in order to transition from the current genomics-centered precision oncology approach to PLPO, a model in which the integration of individual longitudinal, long-term continuous patient monitoring achieves a comprehensive personalized oncology.

Advanced detection algorithms for passive radar (Passive Radar)
Office of Naval Research (ONR); 2022 – 2025

 

 

Space-time sequential schemes and Bayesian inference methods for continuous-time stochastic systems (ALDER)
Office of Naval Research (ONR); 2022 – 2025

 

The aim of this research is to tackle the SDE/PDE/SPDE discretisation (step a) above) and the design of inference methods (step b)) jointly. We claim that, by coupling both procedures, it is possible to improve drastically the performance of the inference algorithms and beat the curse of dimensionality –at least under assumptions that are reasonable in some scenarios.

The main goals of this project are:

1. The design and analysis of computational methods for adaptive discretisation (over timeand space) and Bayesian inference on PDEs.

2. The design and analysis of space-time discretisation schemes and Monte Carlo filters for real-world systems described by partially-observed SPDEs.

3. The design and analysis of a new class of stochastic filters based on deep learning schemes for real-world systems described by partially observed SDEs.

The final goal is, therefore, the design of efficient numerical algorithms that admit a rigorous mathematical analysis in terms of error bounds and their dependence on the dimension of the discrete-time-and-space models.

Patient-centered mental health with attention machine learning (PaC-MAN)
Ayudas para la realización de doctorados industriales 2022; Comunidad de Madrid; 2022-2024

 

The goal of this project, named as PaC-MAN, is to provide eB2 with cutting-edge tools to predict PRMs using explainable deep learning models that combine self-attention and probabilistic methods.

The improvements that PaC-MAN puts forth will have tremendous social and economic impacts. On one hand, fulfilling standard (and lengthy) questionnaires diminishes the time the clinicians can actually spend with patients. Hence, the automatic prediction of PRMs without using such questionnaires translates into spending more time on treating the patients and/or spending those resources on other tasks. Consequently, the patients’ treatment shall improve, which should ameliorate their quality of life. Moreover, it shall also yield shorter recovery times, which implies lower company and social security costs In this regard, this project enhances the mutual benefit between both entities: turning eB2 into a source of tools for patient-centered health and the academic environment into a source of ideas and research results that can be exploited in commercially viable products.

Uncertainty quantification for stochastic physical models: deep filters and space-time Monte Carlo methods (TYCHE)
Agencia Estatal de Investigación (AEI); 2022 – 2025

 

Many phenomena in the physical sciences are represented by dynamic models whose unknown states must be estimated from a collection of related measurements. It is commonplace to describe the evolution of the state using a system of ordinary differential equations (DEs) or partial DEs. Even though these usually reflect well-established knowledge about the behaviour of the physical system, they are still subject to modelling errors and perturbations that escape our control. One way to tackle this issue is by using stochastic DEs (SDEs) or stochastic partial DEs (SPDEs), instead of their deterministic counterparts. Thus, the system is not anymore assumed to evolve in a perfectly predictable way, but some room is made for uncertainty in the form of model misspecification and dynamical noise. This is a more realistic approach to the problem, that has great potential to improve the flexibility and reliability of the models.

In this project we aim at performing estimation, tracking and prediction in complex high-dimensional dynamic systems where uncertainty plays a crucial role. Hence, the latter must be quantified and propagated in a principled way from the initial condition of the system. We tackle this problem from two different angles: Bayesian signal processing and deep learning.

Explainable deep latent representations for patient-centered mental health (EPiCENTER)
Agencia Estatal de Investigación (AEI); 2022 – 2025

 

Mental illness is one of the leading causes of disability, with an estimated prevalence of 12 % of the entire European population in 2015. To alleviate this problem, health systems have been obtaining a large number of psychiatry-related clinical measures. However, they do not typically consider the issues that affect the patients quality of life (QoL).

Patient-centered health has put forth patient-reported measures (PRMs) as an alternative that actually measures what matters to patients (their QoL). Concretely, there are two main families of PRMs: 1) patient-reported outcome measures (PROMs) focus on the patients perception of their own health; 2) patient-reported experience measures (PREMs) are tools for quantifying the patients’ care experiences with the healthcare. In fact, many international organizations promote the widespread use of PRMs to improve the quality of care. This use of standard questionnaires is even more critical in mental healthcare, which lacks objective biomarkers and is mainly based on self- reported symptoms.

Despite the fact that PRMs have been systematically developed over the past decade, their impact has not been significant as they are typically long forms with many fixed questions that disrupt the clinicians workflow. Hence, EPiCENTER will aim to develop techniques to automatically predict PRMs mostly based on passively-gathered data (using a mobile app). Concretely, we shall focus on three relevant PRMs: WHODAS 2.0, PHQ-9, and GAD-7. Additionally, we shall show that our results go beyond mental health, and consider the assessment of the mental well-being of oncology patients with the ECOG and EORTC QLQ-C30 forms.

Inteligencia distribuída en redes inalámbricas
Atracción de Talento Investigador (CAM); 2021 – 2026

 

The goal of this research project is to design a protocol for autonomous spectrum and caching management among multiple heterogeneous networks operating in a changing environment. This project aims to provide a set of heuristics shown to provide good performance in practice. The project will attempt to optimize various metrics, such as throughput and latency, using a
combination of machine learning and more traditional methods based on well understood mathematical and statistical models. Initially, the research will focus on point-to-point flows but, if
time allows it, the framework will be expanded to include relays and broadcast networks.

 

Generalized and refined asymptotics in finite-blocklength information theory (GRAFIT)
Agencia Estatal de Investigación (AEI); 2021 – 2025

  Shannon’s 1948 landmark work has had a great impact in the development of modern communication systems. Shannon’s channel capacity establishes the largest information rate at which data can be reliably transmitted over a noisy channel by using sufficiently long error-correcting codes. This idea has guided system designers for several decades and, nowadays, there exist a number of code families that achieve the channel capacity or perform very close to it. These codes have been adopted in various modern communication standards.

To achieve the channel capacity, we require sequences of very long or infinite duration. In contrast, future communication systems are expected to satisfy stringent requirements on latency and reliability (e.g., for machine-to-machine communications or Internet of Things) that current standards cannot guarantee. Compared to the asymptotic regime of channel capacity, much less is known about the structure of optimal codes with a finite blocklength. This area has recently bounds have been proposed that are extremely accurate in certain scenarios. These bounds depend on the tail probability of high-dimensional variables, and they can be evaluated only numerically and only for simple channels, while for channels of practical interest their computation becomes intractable. A second research direction is to perform an asymptotic analysis of the finite-length performance of the system. While such an analysis often yields easy-to-compute approximations, it is mathematically involved and becomes cumbersome for practical channels. It is therefore not surprising that most of the work in finite-blocklength information theory has focused on simple channel models, where the numerical evaluation of the nonasymptotic bounds can be simplified and where the asymptotic analysis yields clean simple expressions. As a consequence, finite blocklength has received attention within the information theory community, but its impact on related research areas has so far been quite limited.

This project aims at advancing the state-of-the art of finite-blocklength information theory in multiple directions. In the first direction, using Laplace methods, we will perform a refined analysis of the nonasymptotic performance bounds and develop tools to efficiently evaluate them for arbitrary channel models. This is of great theoretical interest, because the simple channel models considered so far in the literature may have a very specific behavior, so more interesting behaviors could be observed by considering more realistic models. In addition, these methods enable the evaluation of the bounds for more realistic channel models, which in turn provides accurate performance benchmarks that are relevant for communication systems. Thus, in the second direction, we apply the methods developed during the course of the project to practical scenarios. The third direction aims at increasing the impact of the results of the project. To this end, we will develop a web application that computes in real time theoretical performance bounds and approximations for wireless channels and quantum channels of interest. In general, this project will advance the theoretical understanding of finite-blocklength communication systems and will make the theory more accessible to researchers in related research areas.
Métodos avanzados de caracterización de incertidumbre orbital aplicados a la detección y seguimiento de basura espacial
Ayudas para la realización de doctorados industriales 2020; Comunidad de Madrid; 2021-2023.
  The project is focused on the analysis and development of advanced methods for orbital uncertainty characterization applied to space surveillance and tracking of space debris, including the following aspects: modelling of the uncertainty in dynamical models and how it affects the orbit of objects around the Earth, such as Earth’s atmospheric drag or solar radiation pressure; quantification of the uncertainty in those dynamical models by parameter estimation using statistical methods; uncertainty estimation and propagation with advanced methods (for instance, particle filters) based on SST sensors measurements (radars, telescopes, etc.); and collision risk assessment using the estimated orbital uncertainty and its corresponding propagation considering the previous developments.

Completed Research Projects

Information Theory for Low-Latency Wireless Communications (LOLITA)
European Comission, ERC Starting Grant; 2017-2023

 

The design of low-latency wireless communication systems is a great challenge, since it requires a different focus than that which is used in current high-speed data transmission systems.  “The project seeks to establish the theoretical framework necessary to describe the fundamental tradeoffs in low-latency wireless communications,” Koch explained. “This enables the design of novel systems that employ resources such as bandwidth and energy more efficiently.” Current wireless communication systems exchange packets of several thousand bits and include large correction codes to protect them against transmission errors. “What we do is to include additional bits to correct possible errors,” Koch stated. In this way, the reliability of the system is guaranteed (what is transmitted is what is received). However, future low-latency systems will exchange information in a much quicker way (almost in real time) and, hence, exchange packets of only a few hundred of bits (a much smaller size), which requires the design of novel correction codes of a shorter length.

Sistema de atención domiciliaria inteligente e interactivo para la mitigación de la pandemia del COVID-19 (IntCARE-CM)
Ayudas para la realización de Proyectos de I+D en Materia de Respuesta a Covid-19, financiados por el FEDER-Recursos REACT- UE; Comunidad de Madrid; 2020-2022
.

 

The project focuses on technological and biomedical field for the care of COVID-19 patients and proposes the development of a comprehensive care and home care system based on the use of artificial intelligence and personalized medicine for early identification and monitoring of COVID and persistent COVID patients.

IntCARE integrates, using mobile phones, wearables, tablets and other devices of use in health:

1) Technologies for automatic monitoring of physical, emotional, social and quality of life from the interaction and use of digital devices and the fusion of multiple sources of information.

2) Accessible interaction and communication technologies for screening, triage and follow-up of patients using voice, processing of the natural language and multimodal interaction.

3) Robotic technologies for assistance and physical, neurological and functional rehabilitation.

4) Visualization and fusion technologies of information for health personnel.

Thus, a holistic and multidisciplinary solution based on care is configured, customized and fully operational to be deployed in pandemic situations, adjusting to the care needs of the population and which will be validated in two clinical studies in hospitals in the Community of Madrid.

Machine learning and massive computation for personalised medicine and quantitative climate analysis (CLARA)
Retos Investigación 2018. Ministerio de Ciencia, Innovación y Universidades.
  In this project we aim at devising classes of dynamical probabilistic models, with allied computational inference methods, which can be used to solve real-world problems in personalised medicine and quantitative climate prediction. While these two fields may look far apart, the key issues to be addressed in terms of model learning and computational inference are of the same kind. We advocate a common methodological approach to problems in both areas and expect a considerable degree of cross fertilization, with ideas and techniques that appear in one field and then can be successfully exploited in the other.
Psiquiatría Computacional y Modelos Integrales de Comportamiento (PRACTICO-CM)
CAM. Consejería de Educación e Investigación; 2019-2022
  Human behavior is understood in many different ways from different fields and sciences. A first notion of behavior has to do with the physical actions carried out by a person in a certain context, a framework within which we would include mobility and other physical activity. On a second level, people, as belonging to an ultra-social species like ours, interact with each other in a social context. Finally, for a  sychologist or psychiatrist, behavior, and especially its alterations, are linked to manifestations of mental disorders, which are usually studied with reference to behavioral patterns considered «normal» in a certain sense. The project is based on the hypothesis that these three visions of human behavior are the projection onto different domains of the same entity, and therefore there is a connection between them that allows explaining and predicting to a certain extent what is  observed in one domain from the others. Our goal is to test this hypothesis and, above all, to advance its application by means of a multidisciplinary approach and team.
Machine Learning Frontiers in Precision Medicine (MLFPM2018)
European Commission Research Executive Agency; 2019-2022
  The goal is to exploit the insights for Precision Medicine, which hopes to offer personalized preventive care and therapy selection for each patient. A technology with transformational potential in analysing this health data is Machine Learning. Machine Learning strives to discover new knowledge in form of statistical dependencies in large datasets. Machine Learning is key to making the vision of Precision Medicine a reality. To meet this challenge, Europe urgently needs a new generation of scientists with knowledge in both machine learning and in health data analysis, who are extremely rare at a global scale. Our ETN’s goal is to close this gap, by bringing together leading European research institutes in Machine Learning and Statistical Genetics, both from the private and public sector, to train 15 early stage researchers. These scientists will help to shape the future of this important topic and increase Europe’s competitiveness in this domain.

Machine learning prediction of the emotional state after the COVID-19 pandemic in depressive psychiatric patients (ESPECTRO)
Ayudas para la realización de doctorados industriales 2020; Comunidad de Madrid; 2021-2023
.

  The goal of this project, named as ESPECTRO, is to provide eB2 with cutting-edge tools to combine and learn from multi-modal information. On the one hand, by providing classical methods with larger expressivity using BNP/ARD priors combined with NVI. On the other hand, by limiting the excessive representation ability of deep probabilistic models using automatic relevance determination priors. Handling heterogeneous data types, including time series, is a must in both cases. This machinery will be tested within an on-going research project of eB2 that aims at finding medically-interpretable models able to predict the emotional state and depression risk of psychiatric patients under current treatment.
Creación de un algoritmo que caracterice el comportamiento humano mediante agregación de datos (Deep-Darwin)
Ayudas Fundación BBVA a Equipos de Investigación Científica 2018; 2019-2021
  The objective of this project is the creation of an algorithm that characterizes the behavior of people through the aggregation of data on a large scale to know their mental state and to be able to help patients who are in psychiatric treatment in a more efficient way. To achieve this, they will collect data from psychiatric patients – who have previously given their consent and guaranteeing their privacy – in collaboration with the Fundación Jiménez Díaz University Hospital. The impact that this investigation can have is to allow an evaluation of a patient’s condition automatically and passively, meaning that the patient does not have to do anything. The psychiatrist can know how the behavior of this person outside the medical consultation and, for example, to be aware of their reaction to a medical therapy to see if it works or there is a change in pattern and from there take the decision you consider timely.
Detección Precoz de Crisis en Autismo (PETRA)
Ayudas a Proyectos de Investigación de tecnologías accesibles Indra-Fundación Universia; 2019-2020
  Patients with autism spectrum disorder (ASD) often present difficult behaviors to manage such as increased anxiety, aggression or sleeping disorders. These behaviors can lead to an increase in medical appointments, greater pharmacological prescription and a worsening in the quality of life of these patients and their caregivers. The main objective of this project is to create a free mobile application, called eB2-ASD (Evidence-based Behavior for ASD) capable of characterizing the behavior of a patient with ASD and find characteristic and personalized patterns that allow predicting episodes of crisis in advance, generating in real time and automatically a direct alert to caregivers on their mobile devices.
Uncertainty propagation meeting space debris needs
European Space Agency; 2019-2020
 
The overall goal of this work is to survey the literature on orbital uncertainty propagation (UP) methods, devise and assess new algorithms where needed, and implement a prototype for the efficient propagation of orbital uncertainties that covers all steps from initialization to the computation of a variety of specific outputs, including collision probabilities and probability distributions for re-entry times. The research is organised around 5 tasks:
  • Analysis and assessment of uncertainty propagation methods
  • Mapping of uncertainties into collision probabilities and re-entry times
  • Design of an end-to-end processing scheme
  • Prototype implementation
  • Prototype qualification and tests
Advanced Bayesian computation methods for modeling and inference in complex dynamical networks (BAYTREE)
Office of Naval Research (USA); 2019-2022
  Complex models, involving many subsystems that interact in non-trivial ways, appear to be ubiquitous in some of the most active fields of science and engineering. There are many difficulties yet, however, both to understand the relevant structures and schemes and to implement useful and reliable algorithms. Our first goal is to investigate a class of dynamical network-like models with layered structure. The goal is to rigorously establish the family of time series models that multilayer network structures can embed. The other class of models we intend to study includes dynamical systems which display features on very different time or space resolutions. Finally, the third aim of the project is to devise algorithms for learning, estimation and prediction that run efficiently on the models of interest. We expect that our approach, based on the joint design of the models and their associated inference algorithms, will bring improvements in accuracy and reliability for range of inference problems on complex systems.
Caracterización automática de comportamiento mediante modelos latentes basados en redes profundas
Ayudas para la realización de doctorados industriales 2018; Comunidad de Madrid; 2018-2021
  The characterization of the patient’s behavior will be translated into the analysis of the time evolution of its latent projection. Sudden changes of the same can be associated to variations of behavior. In a psychiatric patient monitoring (one of the pilot tests contemplated in this project), this may be linked, for example, to a manic crisis in patients with schizophrenia. Given the enormous amount of gathered information for each patient, solutions based on generative models constructed from deep neural networks are proposed. Incorporating the heterogeneity of the behavioral data in this type of models constitutes the first objective of the project. Then, the specific design for the characterization of human behavior with data collected by eB2 in real patients is the next objective. Interpretability in generative models will be emphasized using hierarchical structures and models based-on nonparametric mixtures. The achievement of the proposed objectives will provide the company with tools to support the monitoring and treatment of psychiatric patients that exceed all available solutions worldwide.
Alineación de dominio y curación de datos con redes neuronales profundas. Doctorado Industrial
Ayudas para la realización de doctorados industriales 2019; Comunidad de Madrid; 2019-2022
 
 
The goal of this project is to provide eB2 with cutting-edge tools to perform large- scale joint data wrangling and data aggregation that combines in a principled way recent breakthrough advances in probabilistic modelling with neural networks (the so-called Deep Generative Models or DMGs) and domain alignment. We will design DGMs that specifically tackle the data wrangling problem in all its dimensions, including the ability to tackle the kind of heterogeneous datasets that are expected in big data applications, in which we want to explain away and fuse together multiple information sources of very different nature (sequential data, images, text, social data, …).
 
Change-point Management (CAIMAN)
RETOS 2017; Ministerio de Economía, Industria y Competitividad. 2018-2020
Detection of behavioral changes and its application in psychiatry
Ayudas para la realización de doctorados industriales 2018; Comunidad de Madrid 2019-2022
  The main objective of this project is to detect behavioral changes (daily routine changes) in psychiatric patients using solely the data acquired by the smartphone. Interestingly, these behavioral changes can be used as proxies for (possible) clinical changes. That is, a change in the social activity of a patient with bipolar disorder could be an indicator that he/she moved to the depressive phase. Mathematically, this task can be formulated as the detection of a change point, which is a statistical problem that dates back to the 1940s. Nevertheless, most of the CPD algorithms available in the literature are not appropriate for this application. In particular, to achieve this ambitious goal, we must face some new challenges raised by the nature of the data.
  The goal of the CAIMAN project focuses on the development of online change-point management algorithms, and their application to three selected modern applications. Thus, CAIMAN aims to advance in two different but complementary directions. On one hand, in order to address the aforementioned challenges, the techniques to be developed must go beyond the typical assumptions and simple models considered in classical change-point detection theory. On the other hand, the new framework of change-point management encompasses not only the detection of change points, but also the smart design of the measurement process to optimize performance (change-point sensing).
Micro-fundamentos del comportamiento: Un enfoque basado en las TICs para entender el comportamiento humano y la interacción (aMBITION)
Ministerio de Economía y Competitividad; 2018-2020
  The project is based on the hypothesis that these three visions of human behavior are the projection onto different domains of the same entity, and therefore there is a connection between them that allows explaining and predicting to a certain extent what is observed in one domain from the others. Our goal is to test this hypothesis and, above all, to advance its application by means of a multidisciplinary approach and team. To this end, we will construct models of physical behavior based on the fingerprint of individuals, their strategic behavior based on data from experiments designed specifically for this purpose, and behavior in mental disorders.
Finite-length iterative decoding: fundamental limits, practical constructions and inference (FLUID)
RETOS 2015;
Ministerio de Economía y Competitividad; 2016-2019
  This project aims at building an ambitious theoretical framework for iterative approximate inference with focus on the finite-length regime. Specific project contributions are the following. First, the theoretical characterization, in terms of tradeoffs between rate, block length, and error probability, of short-length transmission under iterative decoding. Second, original GLDPC coding schemes under state-of-the-art decoding to approach these limits. Third, novel techniques to improve approximate inference in iterative decoders and detectors. And fourth, comprehensive experimental scenarios and toolboxes to evaluate code performance as a trade-off between computational complexity and gap to capacity limits, including realistic implementation constraints.
Advanced Bayesian computation methods for estimation, prediction and control in multi-sensor complex system (ADVENTURE)
RETOS 2015; Ministerio de Economía y Competitividad; 2016-2019
  We have recently witnessed the advent of new technologies that hold promise of great improvements to the well-being of elderly people and individuals who suffer from a number of health conditions. Unfortunately, hardware technology alone is not enough to bring all that potential into reality. It guarantees fast and inexpensive access to a wealth of data —yet how to extract knowledge from it and how to make informed and useful decisions is a problem of a different nature. There is an exacting demand of models that impose structure on the data, in order to interpret it, and algorithms that combine those models and the data bunch to estimate key magnitudes, detect ongoing conditions or predict future events. In order to meet these demands, we advocate a Bayesian approach to statistical inference and learning, which encompasses the tasks of model design, comparison and validation, as well as the development of flexible algorithms that fully exploit the features and structure of the underlying models.
Annomalous human behavIour Detection (AID) Explora2014; Ministerio de Economía y Competitividad; 2015-2018
  Brain disorders represent an enormous disease burden, in terms of human suffering and economic cost. In the AID project we will focus on two of them of higher prevalence: schizophrenia and affective disorders (depression and bipolar disease). The aim of the AID project is to explore the feasibility of a method for detecting automatically the behavioral change in the beginning of a relapse of schizophrenic or affective disorder patients in ambulatory conditions by using inertial sensors. The desirable characteristics of this method are: 1) to provide interpretable information, 2) to be easy to personalize, 3) to be able to detect on-line behavioral changes, 4) to include the circadian and the calendar influence on the behavior, 5) to be robust, and; 6) to have a low complexity implementation. In AID we propose to develop a Bayesian on-line changepoint detection method over the sequence of activities provided by a human activity classifier that fulfill all the above requirements.
A new sequential Monte Carlo framework for tracking of nonlinear complex dynamical systems Office of Naval Research; 2015-2018
  In this project, we advocate the development of a new particle filtering framework that still has sequential Monte Carlo integration at its core but is endowed with a number of features that address directly the key issues of dimensionality and complexity. Such features include the partitioning of high-dimensional state spaces, the prevention of the degeneracy phenomenon in importance samplers and the ‘automatic stabilization’ of the tracker. We aim at developing both the methodological and the theoretical aspects of the new framework, and to apply the resulting algorithms to selected problems related to the tracking of multiple and/or complex targets.The design of new and efficient nonlinear trackers for multiple and/or complex targets is relevant to several focus areas of the US Naval Science & Technology Strategic Plan. We will specifically address the application of the new methodology to two problems: the joint tracking of a large number of targets and the forecasting of complex meteorological phenomena for tactical planning.
Intelligent Systems: Concepts and Applications (CASI-CAM-CM) Comunidad de Madrid; 2014-2018
  This project preempts the separation in the knowledge on Intelligent Systems (IS)/ Machine Learning (ML)addressing some of their most important areas in an integrated and cooperative way, together with, necessarily interdisciplinary, applications. The following general objectives are pursued:

 

  • make significant progress in these fields;
  • derive new techniques and concepts through the combination of their perspectives;
  • build an integrated framework able to become established and expanded, providing a competitive international position;
  • get into motion some innovative practical applications including other specific knowledge in order to increase the chance of success;
  • include a scientific and media diffusion plan to obtain the professional and social understanding and appreciation.
Overhead-Throughput-Optimal Signaling Schemes for Next Generation Wireless Networks (OTOSiS) Retos2013; Ministerio de Economía y Competitividad; 2014-2016
  This project advocates a paradigm shift in the way control information is treated in wireless networks. Our philosophy is to view the amount of overhead due to control information as a crucial metric to assess the optimality of physical layer schemes, rather than just an accessory. We will investigate the cost of acquiring network knowledge in dense heterogeneous networks from a fundamental perspective, taking the latency constraints associated with different traffic typologies into account. By studying the information-theoretic limits of wireless networks, we will be able to describe their fundamental overhead-throughput-latency tradeoff. Using these limits, system designers will be able to perform a global wireless network optimization, thereby achieving unparalleled throughput and energy efficiency. We will further propose physical layer signaling schemes that optimally trade overhead, throughput, and latency.
Towards and Efficient Mobile Internet (MobileNET) European Comission; Marie Curie Career Integration Grant; 2013-2017
  It is expected that, very soon, the Internet will connect billions of mobile device users. This places high demands on the communications infrastructure and on the mobile devices. To explore how to use the resources in future communication networks in the most efficient way, we will study the information-theoretic limits of communication networks and suggest communication strategies that attain those limits. We will derive realistic fundamental limits by including asynchronism, noncoherence, and limited codeword length in the analysis. A related topic addressed in this project is the design of mobile devices. Using tools from information theory, we will study the fundamental tradeoff between performance, robustness against nonlinearities in the devices, and implementation complexity, aiming at novel encoding and decoding algorithms that can be implemented in hardware.
Computational Inference in High Dimensional Random Complex Systems (COMPREHENSION) Ministerio de Economía y Competitividad; 2013-2015
  The term «complex system» is often used to describe a network of elementary units whose collective behavior depends not only on the features of these constituent blocks but also, and specially, on their interactions. In this project, we investigate dynamic, high-dimensional and random complex systems and we aim at developing new methodologies for computational inference which are both theoretically sound and practically effective in this setup.While the advance in the theoretical and methodological field is of utmost importance, we also pursue practical applications of the new methods. The most ambitious goal is the modeling of atrial fibrillations (AF) in the human heart; we also investigate relevant problems related to wireless communications and sensor networks (WCSNs), including collaborative routing and distributed implementation of statistical signal processing methods on multi-hop WCSNs. The third axis on which we move from theory to applications deals with environmental applications.
Advances in Learning, Communications and Information Theory (ALCIT) Ministerio de Economía y Competitividad; 2013-2016
  With the current technology trends, communication networks are evolving towards ever-complex systems consisting of a large number of heterogeneous nodes that enjoy enhanced capabilities for sensing, storing, processing and transmitting data in many sophisticated forms. This project deals with two important aspects of the design and analysis of these networks. We first study the information-theoretic limits of the aforementioned networks. Secondly, we need to learn from the data captured from these devices that are constantly monitoring a changing, diverse and complex environment. Our main goal is to advance towards the solution of a number of fundamental problems that arise in this scenario, both in terms of new formal methodologies and numerical techniques and in the demonstration of their validity by means of an adequate hardware platform.We will apply the obtained results in machine learning and finite length information theory to solve all relevant psychiatric problems: the remote registration of patient’s activities. We will build two demonstrators using currently available hardware and software elements as an intermediate stage to the building of the final system.
Machine Learning for Personalized Medicine (MLPM) European Comission; Marie Curie Actions; 2013-2017
  MLPM is a Marie Curie Initial Training Network, funded by the European Union within the 7th Framework Programme. MLPM has started on January 1, 2013 and will be carried out over a period of four years. MLPM is a consortium of several universities, research institutions and companies located in Spain, France, Germany, Belgium, UK, and in the USA. MLPM involves the predoctoral training of 14 young scientists in the research field at the interface of Machine Learning and Medicine. Its goal is to educate interdisciplinary experts who will develop and employ the computational and statistical tools that are necessary to enable personalized medical treatment of patients according to their genetic and molecular properties and who are aware of the scientific, clinical and industrial implications of this research.
Environment and Genes in Schizophrenia (AGES) Comunidad de Madrid; 2012-2016
  Mental disorders, as a whole, are the leading cause of disability world-wide. Schizophrenia has been described as the most devastating psychiatric disorder with a prevalence of 1% in the general population. It is estimated that more than 50.000 people in Madrid experience schizophrenia during their lifetime. The AGES-CM consortium (Ambiente y Genes en Esquizofrenia – Grupos de Investigación de la Comunidad de Madrid) is formed by leading schizophrenia research groups from the Comunidad de Madrid (CM). These research groups from our Community lead the field of schizophrenia research in Spain and at the international level in many aspects of the disease.
New Computational Inference Methods for Spatial Dynamical Models Ministerio de Educación, Cultura y Deporte; 2012-2013
   
Estimation, Transmission and Optimization in Sensor Networks (ETORS) Comunidad Autónoma de Madrid and Universidad Carlos III de Madrid; 2011
   
Analysis, Design and Optimization of Next Generation Wireless Communications Systems Ministerio de Ciencia e Innovación; 2010-2012
   
Distributed Learning, Communication and Information Processing (DEIPRO) Ministerio de Ciencia e Innovación; 2009-2012
   
Ubiquitous Service Arquitecture for the Mobile Super Prosumer (uSERVICE) Ministerio de Industria, Turismo y Comercio – Plan Avanz@; 2009-2010
   
Foundations and Methodologies for Future Communication and Sensor Networks (COMONSENS) Ministerio de Ciencia e Innovación (Consolider-Ingenio 2010); 2008-2014
   
Intelligent Intermodal Freight Transport (TIMI) CDTI (programa CENIT); 2007-2010
   
Smart Monitorization (MONIN) Ministerio de Educación y Ciencia; 2007-2009
   
Consortium for the Development of Advanced Technologies for Medicine (CDTEAM) CDTI (CENIT Consortium); 2006-2010
   
Approximate Inference for Communications Marie Curie Outgoing Fellowship (FP6 – European Union). 2006-2009
   
Multimedia Distributed Processing (PRO-MULTIDIS-CM) Comunidad Autónoma de Madrid; 2006-2009
   
Eficient Multimedia Communications Enabled by Advanced Learning Algorithms (CREMA3) Ministerio de Educación y Ciencia; 2006-2008