CiTIUS: 8 challenges, 8’: meet our research in less than 10 minutes

NL4XAI – ‘Interactive Natural Language Technology for Explainable Artificial Intelligence’

CiTIUS leads the first European Training Network (ETN) on Natural Language and Explainable Artificial Intelligence, which brings together 18 partners from 6 different countries. The main goal of this network, funded by the Horizon 2020 programme, is to produce intelligent machines able to explain their behaviour and decisions through interactive explanations in Natural Language, just as humans naturally do.

José María Alonso Moral

Real-time processing of multi and hyperspectral images

At CiTIUS we develop solutions linked to real-time image processing of remote sensing data, with special interest in multi and hyperspectral images. Together with other more classical approaches, some techniques such as Deep Learning or Convolutional Neural Networks have helped us to achieve efficient solutions adapted to high performance computing platforms.

Dora Blanco Heras

TRAFAIR – ‘Understanding Traffic Flows to Improve Air Quality’

TRAFAIR is a European project aimed at developing innovative low-cost infrastructures and services for air-quality monitoring and forecasting at urban scale in European smart cities. Physical simulation and AI-based models are combined to generate spatio-temporal predictions of air pollutant concentrations from emission sources (specially traffic flow) and meteorological conditions.

José Ramón Ríos Viqueira

Applying Deep Learning to biomedical image analysis

Our group at CiTIUS is obtaining great results through the application of Deep Learning architectures to panoramic X-ray images, in order to estimate chronological age and gender. We have published the first full automatic system to estimate the chronological age. At present we are working on the application of Deep Learning techniques to the early detection of covid-19 in X-ray radiographs.

María José Carreira Nouche

Language Technologies and Big Data

We are particularly interested nowadays in analysing texts written by people on the Internet, mostly on Social Media. We want to understand aspects such as the credibility of information, emotions, or sentiments. One of our remarkable results from CiTiUS in this field is eRisk, a worldwide initiative to foster research on the interactions between natural language use and psychological problems. We have been organizing this campaign-style initiative since 2017.

David E. Losada

MENELAOS NT – ‘Multimodal Environmental Exploration Systems – Novel Technologies’

MENELAOS project is funded by the European Union’s Horizon 2020 programme through a Marie Curie action. Our goal is to apply novel technologies in the area of multi-sensor data fusion, in order to optimally combine the information delivered by different sensors. One of our recent achievements at CiTIUS has been to design and manufacture a chip with a built-in photovoltaic energy harvesting system, with the ability to auto-start from 3 orders of magnitude lower than current commercial solutions

Paula López

Artificial Intelligence to improve the detection and tracking of objects in videos

We apply Artificial Intelligence techniques to computer vision, focusing on the detection and tracking of objects with Deep Convolutional Neural Networks. Our group at CiTIUS has designed a deep Convolutional Neural Network to track multiple objects in video. The main novelty is that this network can track accurately up to 60 simultaneous objects in real time.

Manuel Mucientes

Víctor Brea

Glocal Machine Learning for smart devices

During years we have carried our research at CiTIUS on continual learning from environment interaction, especially when working with robots. Currently we are focused on glocal learning, thus addressing asynchronous federated continual learning with non-stationary data. We mostly focus on situations where data is partially labelled, non-identical among the participants, and with an underlying distribution that changes over time in unforeseen ways, causing what are known as concept drifts. The local devices are able to detect when this happens, so that they can determine when to learn and from which data. This allows the process to be asynchronous, scalable, and able to deal with the catastrophic forgetting, still preserving data privacy.

Roberto Iglesias

Xosé Manuel Pardo