Research Projects

Ongoing Research Projects

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.
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.
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.
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.
Psiquiatría Computacional y Modelos Integrales de Comportamiento (PRACTICO-CM)
CAM. Consejería de Educación e Investigación; 2019-2021
  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.
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, …).
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.
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.
Change-point Management (CAIMAN)
 RETOS 2017; Ministerio de Economía, Industria y Competitividad. 2018-2020
  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.
Information Theory for Low-Latency Wireless Communications (LOLITA)
European Comission, ERC Starting Grant; 2017-2022
  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.
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.

Completed Research Projects (2008-2018)

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