DaSCI Technology Applications Area
Cutting-edge research in Data Science and Computational Intelligence, and one of the main objectives of the DaSCI Institute is to offer applied research. The following research area describes the reality of knowledge transfer from the university to other social spheres. Backed by a large number of projects, research contracts with companies and international publications, the applications developed in DaSCI are the following:
Research lines in the area of DaSCI Applied Technologies
AI Applications in Civil Engineering
The most advanced countries are facing the onset of the 4th industrial revolution, brought about by technologies such as artificial intelligence, advanced monitoring and robotics. The Construction and Civil Engineering industry will be part of this revolution, due to the relatively low cost of digitalization technologies in relation to the huge operation and maintenance costs of infrastructures. The amount of data and information in real time coming from monitored infrastructures is expected to increase exponentially in the coming years. This information has the potential not only to reduce the national budget on infrastructure maintenance by billions, but also to dramatically change the way the 21st century infrastructures will be designed, built and operated. In this context, the overall objective of this research area is to contribute to this revolution and to academically lead the development of innovative computational intelligence technologies for the construction industry of the 21st century.
More specifically, the following research lines are currently being carried out at DaSCI:
- Development of digital twins and cyber-physical systems for intelligent management of structures and infrastructures.
- Development of intelligent systems for remaining useful life prognostics of infrastructures.
- Development of technologies for robotic maintenance of infraestructures.
- Intelligent systems for the design, planning and construction of infrastructures.
- Intelligent systems for failure detection and predictive maintenance of infrastructures.
Knowledge extraction from scientific and technological sources
Currently, innovation, development and research activities, and in particular science and technology, generate large amount of information derived from the scientific method, such as articles, theses, patents, clinical trials, among others. In fact, the total volume increases year by year, following a constant growth pattern.
In this sense, the global system of science and technology can be understood as a complex structure in which academic and industrial actors, and the global knowledge generated, relate to each other, and can be represented as a network or graph. Thus, there can be nodes related to people (researchers), institutions, countries, documents, concepts, journals, etc. Between these, different types of relationships can be established. Furthermore, in this network, relationships are heterogeneous and dynamic, i.e., relationships are not only established between two nodes of the same type, but maybe involved in different types of nodes, and these may change over time.
Traditionally, the extraction of knowledge from this type of data has been developed using a bibliometric perspective. But now, it is necessary to design and develop new models, algorithms and software tools capable of extracting hidden knowledge from large volumes of information with a multitude of heterogeneous and dynamic relationships between its components employing techniques and method from artificial intelligence, machine learning and complex systems.
AI and remote sensing for biodiversity and global change
Global change and the loss of biodiversity are one of the major challenges that threat human well-being at present. Detecting, understanding and monitoring spatiotemporal changes in the earth surface from the local to the global scale are of paramount importance for researchers and policy makers all around the world. Currently, one of the main resources for conducting such studies is remotely sensed data, from ground-based sensors to aerial and satellite imagery. Daily, Terabytes of data are produced at different spatial, spectral, radiometric and temporal resolutions. Extracting useful patterns from this type of data is very complex. Artificial intelligence algorithms in general and deep learning models in particular have a high potential to detect complex patterns in this type of data. Our research in this topic is fusing remote sensing and IA technologies with the aim of contributing to the Sustainable Development Goals.
Internet of Things and Industry 4.0
The introduction of new technologies in the industry is changing the way many companies work. Technologies such as the Internet of Things (IoT), cloud services, big data, robotics, artificial intelligence, 3D printing, or Blockchain will be the key to how companies offer their products and services in the future.
What has been called Industry 4.0 is defined as an advanced production ecosystem, automated and interconnected thanks to these technologies, and especially thanks to IoT, understood as the connectivity of intelligent devices so that they can detect and communicate to collect real world data easily. The goal of Industry 4.0 is to actively drive the reshaping of the industry by implementing disruptive technologies that allow manufacturing equipment to be connected to a web-based network and gain substantial value.
These advanced technological solutions will facilitate the management and analysis of information in real time, as well as decision making, allowing the creation of improved systems for production.
Examples of the implementation of IoT in Industry 4.0 are intelligent agriculture, waste collection systems or monitoring of goods.
Currently the biggest challenge in these areas is to establish what data to collect and how to collect it, how that information will be analyzed to obtain useful information and thus make the right decisions. Additionally, all processes must be carried out in secure environments to avoid cyber threats in order to keep data, systems and networks protected.
Translational biomedical research
The rapid development and implementation of data repositories that contain representations of complex objects, such as time series or spatial data, has not been accompanied by a greater availability of tools that allow the search of these databases in terms close to the language of and based on criteria that are meaningful to the intended users of those data collections. These shortcomings are evident in particular fields such as biology or medicine, where a large amount of data accumulates. However, retrieved knowledge often frustrates the expectations of interested users because it consists of familiar, homogenized patterns or long and intractable descriptions. Therefore, we believe that one of the challenges in the post-genomic era is to extract useful knowledge and, more importantly, to transform this knowledge into actionable knowledge. That is, deciphering the available information and converting it into hypotheses that can predict new knowledge that would eventually improve the decision-making systems that affect our lifestyle. Our biomedical research has been based on multiple measurements including but not limited to multiple imaging, genetics, molecular biology, multi-omics, clinical diagnoses and comorbid phenotypes, phenomics, real-time physiological signs, EHR and EMR, etc. The different target diseases include dementias, mental disorders, cardiovascular diseases and various types of cancer, among others. Our challenge is to generate novel artificial intelligence and machine learning tools that integrate the data available in knowledge bases utilized by regular as well as AI clinicians. The total or partial coincidence with interconnected pre-existing knowledge bases will reduce the universe of possible diagnoses of a patient and, consequently, will shorten the time that the patient needs to be treated. A decision-making performed by an AI clinician will help identify the treatments that are best suited to diagnosis and patient follow-up.
Educational Data Mining
Educational Data Mining (EDM) and Learning Analytics (LA) are research areas that aim to develop methods for measuring, collecting, exploiting, and analyzing data from educational environments to solve critical educational issues and problems. This research area, which began to be developed at the beginning of 2000, has grown enormously, successfully addressing student modeling problems, predicting academic performance, personalizing teaching, detecting abnormal behavior, recommending resources, or automatically constructing courses. Some of the most recent interest areas are related to improving the models obtained in terms of transferability, effectiveness, interpretability, applicability, and generality. On the other hand, with the appearance of new teaching systems that make use of multiple methodologies, there is a growing interest in the use of multimodal data, as well as the use of data from environmental and personal sensors that even allow access to students’ brain data, providing information on aspects such as their level of attention and concentration during classes.
Computational business management and marketing models
In the real world there are many examples of complex systems such as the social ones (people have a natural tendency to form groups: families, circles of friends, professional or religious groups, cities, nations, etc.), the business/economic ones (companies, clients, etc.) and the biological ones (e.g. metabolic networks). Given the important role that complex systems play in our environment, their understanding, quantification, prediction and eventually control have acquired a capital importance, becoming one of the main intellectual scientific challenges of the 21st century.
In the field of Economics, these complex phenomena can be modeled through different types of simulation models: classical econometric models, system dynamics models and models based on artificial intelligence (AI) techniques such as social simulation carried out with agent-based models (ABM), among others. These models allow to extract knowledge that helps to understand how the relationships between customers, brands and media drive all market dynamics. Instead of thinking about big ideas and testing them in the market, we can test them by running experiments in a virtual marketplace and learn from those simulations by continuously asking “what-if” scenarios at a negligible cost. In addition, marketers can study how “word-of-mouth” and social influences travel in a network of consumers, thus being able to test the effects of advertising campaigns and marketing strategies on the diffusion of innovation at the macro level.
In this way, the use of ABMs and other AI technologies can improve marketing and business management processes by providing new knowledge to decision makers and supporting decision making on an unprecedented scale. Our main objective in this line of research is to exploit the use of AI/computational intelligence techniques (in particular ABM, social network analysis, interpretable and non-interpretable machine learning algorithms, advanced search and optimization algorithms, and probabilistic and enabling knowledge representation and reasoning frameworks) to design realistic models of marketing/business management problems, develop knowledge extraction tasks from data in these domains, and assist the associated decision maker. In particular, several of our research developments are being applied in the market through a technological alliance with the marketing consulting firm R0D Brand Consultants in the commercial product Zio Analytics, which incorporates our AI solutions for strategic branding and consumer behavior modeling in virtual markets.
Sustainable Development Goals
The Agenda 2030 for Sustainable Development was approved by United Nations Member States at the World Summit for Sustainable Development in 2015. In order to eradicate poverty, protect the planet, and ensure prosperity for all people, 17 goals, 169 targets, and a political declaration were established. Education, equality, access to energy, availability of water, infrastructure development or guidelines for sustainable consumption, are some of the objectives of this international agenda that takes over the eight Millennium Development Goals in force since 2000. It is also universal, and therefore applicable to all countries, including the so-called developed ones.
The 17 Sustainable Development Goals (SDA) aim to be a global instrument to eradicate poverty and reduce inequalities and vulnerabilities, under the paradigm of sustainable human development.
There is a wide range of emerging digital technologies with enormous potential to promote greater quality of life in human beings, whether at the level of economic, food, health, and welfare support. For example, digital technologies such as apps or Augmented Reality, 3D models used in the Digital Twins, 3D Printing, or Virtual Reality, as well as the emerging paradigms of AI such as Deep Learning, Blockchain or Edge Computing.
We work to discover the potential of Artificial Intelligence and digital technologies, to solve the current challenges of Agenda 2030, and therefore those proposed by the 17 Sustainable Development Goals. We propose a set of socio-technological recommendations, by which it is possible to act in support of the achievement of these objectives.
Biosignal and medical imaging processing
Over the last few decades translational neuroscience has transitioned from qualitative case reports to quantitative, longitudinal and multivariate population studies in the quest for defining patterns of disease pathogenesis, prognostic indicators and treatment response. Massive data analysis and computational processing has revealed previously unknown interactions, providing potential biomarkers for diagnosis and prognosis in clinical practice.
The field of biosignal and medical imaging processing consists on using computational and mathematical approaches based on the statistical learning theory with the final objective of developing computer-aided diagnosis systems in neuroscience. In the sense we aim to provide tools for clinicians that support the study and diagnosis of neuropathological diseases, such as Alzheimer or Parkinson diseases; tools that will largely influence treatment and patient management. On the other hand, advanced signal processing methods developed for neurophysiological signals allowed the measurement of the response to specific stimulus and provide the arena for the construction of complex functional connectivity models.
From the perspective of data analysis, a variety of statistical signal processing strategies are proposed, including the statistical comparison of individual’s data with reference images of control subjects for the in-vivo assessment of brain functional/structural parameters, the joint modelling of brain images to understand the different sources of variability in brain function and structure, and many other approaches that may support an informed decision in neurodegenerative diseases. Morever, the use of deep learning and bayesian deep learning based models for image and time series data constitute a powerful alternative to classical analysis methods, not only for discriminative analysis but also for exploratory analysis
Connected health is the provision of health and medical care services using a digital ecosystem. Digital ecosystems have the capacity to obtain and exchange different types of data through multiple sources of information (devices with sensors, medical equipment, medical images, expert knowledge, etc.) and acquire new knowledge with the main objective of making better decisions in health matters. In the connected health scenario, the main actors are citizens, professionals, as well as health centers and services where digital ecosystems provide multiple opportunities for health care and improvement. These opportunities include, for example, reducing time, distance and access limitations to remote locations, providing continuous and real-time monitoring, improving diagnosis and treatment processes, as well as optimizing organizational processes.
Security and video surveillance
The increasing availability of artificial intelligence tools and algorithms for video processing has increased the number of potential applications in the field of security through video surveillance techniques.
In this line of research at the DaSCI Institute, algorithms and methodologies are being developed for video processing in the field of security. We must highlight the design of algorithms for target detection, tracking of people or other targets, or analysis of anomalies in the processing of crowd or flow of people.