Mapas (Máquinas de apredizaje para problemas y aplicaciones singulares)

Organismo: Ministerio de Economía y Competitividad. MINECO

Periodo: 2017-2019

 

Resumen español:

Los 10 grupos de I+D solicitantes (cinco procedentes de la actual Red Temática DAMA, TIN2015-70308-REDT, a los que se suman aquí otros cinco con los que se ha mantenido cooperación demostrable en el ámbito que enfoca esta propuesta) comparten interés activo en temas localizados en las actuales fronteras del Aprendizaje Máquina: muy particularmente, en los denominados problemas singulares; en especial, clasificación desequilibrada, cuya frecuente presencia en importantes problemas reales ha desatado un interés velocísimamente creciente; sin olvidar los muy próximos de clasificación ordinal, microclasificación, clasificación multietiqueta y análogos. Las visiones, métodos y algoritmos con que actualmente se abordan esos problemas adolecen, salvo muy contadas excepciones parciales, de falta de fundamentos analíticos, con lo que sufren de sesgos y generan riesgos de incorrecta aplicación.

La Red Temática para la que se solicita ayuda operará constituyendo grupos mixtos que se centrarán en el intercambio de ideas y la cooperación en la concepción, el desarrollo, la evaluación y la aplicación real de métodos, procedimientos y algoritmos fundamentados para abordar dichos problemas; considerando sus extensiones, como son las formulaciones multiclase y los costes funcionales, y concediendo especial atención a las implementaciones más prometedoras, basadas en diversidad, profundidad y carácter dinámico; así como a versiones del aprendizaje menos estudiadas que el supervisado: aprendizaje por refuerzo y por transferencia.

Para enfrentar en la mejor posición aplicaciones prácticas, se consideran también situaciones en que la información disponible para el diseño es parcial, como valores perdidos, ruido en etiquetas, etc.

Se seleccionan ámbitos de aplicación (Negocio, Salud, Visión Computacional, etc.) relevantes, generales y abiertos para facilitar acuerdos con entidades interesadas para llevar a cabo aplicaciones prácticas e internacionalización de resultados.

 

Resumen inglés:

The 10 R&D teams that present this application (five of them coming from DAMA Net, TIN2015-70308-REDT, plus another five that have collaborated in provable form in the topics that this proposal considers) have a common active interest in frontier-of-knowledge Machine Learning subjects. Namely the teams address the so-called singular problems: In particular, imbalanced classification, which is pervasive in important practical problems, and, consequently, has generated a tremendously increasing interest. Associated singular problems, such as ordinal classification, microclassification, multilabel classification, and similar, also merit the attention of the R&D teams. The perspectives, methods, and algorithms that are being applied to solve these problems, with very few partial exceptions, are not supported by principled analyses. This produces undesirable biases and wrong application risks.

This Network will operate by constituting mixed teams to focus on the interchange of ideas and the cooperation to conceive, develop, evaluate, and practically apply new principled methods, procedures, and algorithms for such singular problems. Their extensions (multi-class formulations and functional costs, for example) will be also studied, and special attention will be paid to the more premising implementations: Ensembles, deep architectures, and dynamic machines. Additionally, less usual (in general terms) learning situations (mainly, reinforcement and transfer learning) will be included. To address from the best position practical applications, partial information cases (missing values, noisy labels, etc.) will be studied, too.

Some relevant application areas (Business, Health, Computer Vision, etc.) are initially selected, keeping them wide and open to make easier to establish agreements with interested entities for carrying out cooperations and practical applications, as well as to internationally expand activities and results.

Diversificación Avanzada de Máquinas de Aprendizaje (Advanced Diversification for Learning Machines)

DIVERSIFICACIÓN AVANZADA DE MÁQUINAS DE APRENDIZAJE (ADVANCED DIVERSIFICATION FOR LEARNING MACHINES) TIN2015-70308-REDT 

Financial Entity: Ministerio de Economía y Competitividad.MINECO
Principal Investigator: Aníbal Ramón Figueiras Vidal. En la UCO: Pedro Antonio Gutiérrez Peña
Budget: 35.000 Euros. 

 

La   Red   Temática   DAMA,   constituida   por   6 grupos   de   investigación   con   demostrable  cooperación   anterior,   se   orienta   científico-­técnicamente   a   introducir   nuevos métodos   de  diversificación,   mejora   de   los   conocidos, combinaciones   de   ellos,   empleo   de   nuevos aprendices (incluyendo  las  muy  celebradas  Redes  Profundas),  extensiones  a problemas  de  Aprendizaje   Máquina   de   particular   relevancia (desequilibrados,   ordinales,   semi-­   y   no-­supervisados,  etc.),  aprovechamiento  en  tareas  secundarias  (como  imputación, selección  de  rasgos  y  ponderación  de  errores  muestrales), llegando  a  las  aplicaciones,  preferentemente  en   los   ámbitos de   Finanzas,   Energía,   Márketing   Inteligente   e   Industria Química,  Farmacéutica  y  Biotecnológica,  con  el  apoyo  de  media docena  de  empresas  especializadas.

Se   operará   constituyendo   equipos   inter-­grupales   dedicados   a   temas   específicos; celebrando   unas   Jornadas   Anuales   de   la   Red   Temática   DAMA   con   propósitos   de   coordinación, formación,   investigación,   transferencia   y   difusión;;   y buscando   la   visibilización internacional   vía   resultados científicos,   propuestas   de   proyectos   europeos,   y   cooperación   internacional   (preferentemente   a   través   de   estructuras análogas,   como   el   “Data   Science   &  Engineering   Consortium”  USA-­LatAm-­España,   del   que   dos   universidades   solicitantes  ya   forman  parte). 

Diversificación avanzada de máquinas de aprendizaje (Red DAMA)

Organismo: Ministerio de Economía y Competitividad. MINECO

Periodo: 2016-2017

 

Resumen español:

La Red Temática DAMA, constituida por 6 grupos de investigación con demostrable cooperación anterior, se orienta científico-­técnicamente a introducir nuevos métodos de diversificación, mejora de los conocidos, combinaciones de ellos, empleo de nuevos aprendices (incluyendo las muy celebradas Redes Profundas), extensiones a problemas de Aprendizaje Máquina de particular relevancia (desequilibrados, ordinales, semi y no supervisados, etc.), aprovechamiento en tareas secundarias (como imputación, selección de rasgos y ponderación de errores muestrales), llegando a las aplicaciones, preferentemente en los ámbitos de Finanzas, Energía, Márketing Inteligente e Industria Química, Farmacéutica y Biotecnológica, con el apoyo de media docena de empresas especializadas.

 

Resumen inglés:

Six research groups that have (documented) previous cooperation experience, would compose the DAMA Net. The scientific and technical contribution of that net is towards diversity in Machine Learning, looking for new techniques and improving those already available, plus introducing new learners (event the much celebrated Deep Networks). Extensions of the results to particular relevant Machine Learning Problems imbalanced, ordinal, semi- and non-supervised, etc.- and taking advantage of them to address secondary tasks (such as imputation of missing values, feature selection, and sample error weighting). With respect to practical applications, we will focus on Finance, Energy, Intelligent Marketing, and Chemical, Pharmacological and Biotechnological Industry problems. Six companies with recognized presence in these areas will help us in these directions.

The research activity will be carried out by constituting inter-group teams that will concentrate in specific issues. Annual Days of DAMA Net will be celebrated and will include coordination, educational, research, technology transfer, and diffusion components. International visibility will be obtained with scientific results, EU project proposals, and international cooperation (mainly via similar nets, such as the USA-LatAm-Spain Data Science and Engineering Consortium, which already includes two of our Universities).

Algoritmos de clasificación ordinal en energias renovables (ORdinal Classification and prediction Algorithms in Renewable Energy, ORCA-RE)

ALGORITMOS DE CLASIFICACION ORDINAL Y PREDICCION EN ENERGIAS RENOVABLES     (ORDINAL CLASSIFICATION AND PREDICTION ALGORITHMS IN RENAWABLE ENERGY, ORCA-RE) TIN2014-54583-C2-1-R 

Financial Entity: Ministerio de Economía y Competitividad.MINECO
Principal Investigator: César Hervás Martínez and Pedro Antonio Gutiérrez
Budget: 79.200 Euros. 

 

Given that fossil energy resources will not satisfy the energy demand of the world population within a relatively short period of time, a very important research trend is now investigating in alternative sources of efficient, reliable and clean energy, to boost the performance of current infrastructures. Predicting the amount of energy produced is essential for assuring an effective inclusion of these energies in the electrical network. This kind of energies are associated to physical phenomena with an important unknown random component, producing high variability. In this way, their prediction is not affordable, in general, using classical predictive methodologies. Because of this, the coordinated project ORCA-RE is aimed to explore, develop and extend various machine learning methodologies to tackle the problem of production estimation of Wind Energy, Solar Energy and Wave Energy. Different paradigms, such as ordinal classification or time series segmentation, will be analysed, which have a great interest for the central problem of this project and which have been scarcely studied in comparison to nominal classification, both in Spain and in the rest of the world. Both research groups have previous, coordinated and contrasted experience in this field. The predictions based on ordinal classification will be compared to the use of segmentation, to evaluate the influence of the temporal component. In this way, we are pursuing the following objectives related with the challenge of alternative energy sources:
1) To use Computational Intelligence techniques to develop new ordinal classification models and imbalanced ordinal classification models, and to analyse new classifier evaluation metrics. Use of mono and multi-objective hybrid methodologies (this last paradigm needed because some of the classifier evaluation metrics are opposite).
2) To develop time series segmentation algorithms applied to renewable energies based on statistical and bio-inspired methods. Medium-term prediction using the result of these segmentation algorithms.
3) To develop new bio-inspired models for the evolution of classifiers and regressors using grouping genetic algorithms and new single population co-evolution models using the Coral Reefs Optimization paradigm.
4) To apply these models to different renewable energy problems, mainly prediction and resource estimation in Wind Energy, Solar Energy and Wave Energy. Application to other real problems of time series segmentation and prediction.
5) To develop a software package to be used in the framework NNEP incorporating all the new models developed in the project, and a package for the WEKA framework incorporating some of the main ordinal classification methods, which would allow the spreading of this paradigm in the scientific community. 

Climate tipping points: Detection and analysis of patterns using an ordinal regression approach. Proyecto Ariadna.

Financial Entity: European Space Agency (ESA)
Principal Investigators: César Hervás Martínez, Pedro Antonio Gutiérrez Peña

 

 This project aimed at gaining knowledge about the typical statistical behaviour of paleoclimate data series before the occurrence of a tipping point to allow the detection of warning signs for upcoming abrupt climate transitions. Tools developped in the field of machine learning and soft computing were applied to Greenland ice core data to provide insights on the dynamical system under study. 

In this project we proposed a time series segmentation algorithm combining a clustering technique and a genetic algorithm to automatically find segments sharing common patterns from paleoclimate time series. The segments were transformed into a six-dimensional space composed of six statistical measures most of which have been previously considered in the detection of critical transitions.

Experimental results showed that the proposed approach applied to paleoclimate data could effectively assess warning signals of Dansgaard-Oeschger (DO) events and uncover commonalities and differences in their statistical and dynamical behaviour. In particular, strong warning signals were detected in the GISP2 and NGRIP δ18O ice core data for several DO events (e.g. DO 1, 4, 8 & 12) in the form of increasing variance, autocorrelation and mean square error. The increase in mean square error, suggesting nonlinear behaviour, has been found to correspond with an increase in variance prior to several DO events for ∼90% of the algorithm runs for the GISP2 δ18O dataset and for ∼100% of the algorithm runs for the NGRIP δ18O dataset. The proposed approach applied to paleoclimate data provides a novel visualisation tool in the field of climate time series analysis of critical transitions. 

NEuro-MOdelado AVAnzado para Clasificación Ordinal y Nominal mediante algoritmos de aprendizaje híbrido. Aplicaciones en teledetección para agricultura y en biomedicina de trasplantes (NEMO-AVACO)

Junta de Andalucía.Proyecto P2011-TIC-7508

 Financiación: 64.377€

NEMOTECH: Técnicas de Neuro-Modelado utilizando Algoritmos de Aprendizaje Híbridos. Aplicaciones en Biomedicina de Trasplantes, Agronomía y Microbiología Predictiva

Proyecto de Excelencia Plan Nacional I+D+i, MICINN, TIN2011-22794

Financiación: 65.461€

Red Temática Española para el Avance y la Transferencia de la Inteligencia Computacional Aplicada (ATICA)

Ministerio de Economía y Competitividad, Subprograma de Acciones Complementarias a Proyectos de Investigación Fundamental No Orientada, TIN2011-14083-E.

Financiación: 12.000€

Análisis espacial y de clasificación ordinal de la distribución geográfica de enfermedades mentales en Andalucía

Financial Entity: INSTITUTO DE SALUD CARLOS III
Principal Investigator: Garcia-Alonso, Carlos Ramon
Budget: 23440,12 Euros. 

Research on financing systems' effect on the quality of mental health care (Refinement)

RESEARCH ON FINANCING SYSTEMS' EFFECT ON THE QUALITY OF MENTAL HEALTH CARE (REFINEMENT) Code: EC-GA N°261459

Financial Entity: VII PROGRAMA MARCO DE LA UNIÓN EUROPEA, COMISIÓN EUROPEA
Coordinator: Professor Francesco Amaddeo
Person responsable in Spain: Carlos Ramón García Alonso