The large amount of digital information in society is inherently connected to advances in technology, the WWW, the rise of smart phones, cloud services, the Internet of Things. This big data phenomenon is unstructured and transient. Motivated by this reality, the Big Data pattern analysis and modeling group aims at converting raw data into valuable knowledge by uncovering hidden patterns and unknown correlations in order to stimulate scientific discoveries and optimize processes within the society through the understanding of social and collective behavior. Our research, applied to a broad
range of applications, spans from data management and advanced data analysis to service platforms. More specifically, our research interests include:

  • Multi-dimensional data analysis (preprocessing, feature extraction, dimensionality reduction, knowledge discovery)
  • Spatial, temporal (including multi-dimensional time series), spatio-temporal data mining
  • Computational frameworks and modeling (data driven, statistical, probabilistic, graph (time-evolving and static))
  • Data representation, integration, fusion and high-order analysis
  • Data compression
  • Streaming data analysis and decision making
  • Applications of interest: energy management, economics, environment, human activity, social behavior, seismology, neuroinformatics


Biomedical Image analysis

The Section for Biomedical Image Analysis aims to the development of computer-based approaches and decision-support tools applicable to clinical practice. Brain imaging methods like conventional Magnetic Resonance Imaging (MRI), Computerized Tomography (CT), functional MRI (fMRI), Galactography and Mammography are commonly utilized in clinical practice and provide noninvasive images of anatomy and function of living tissues. In our lab, image analysis methodologies focus on medical image segmentation and registration, feature extraction from Regions Of Interest (ROIs), high dimensional pattern classification, and analysis of morphological variability, such as vessel anatomical tree-shape structures. Clinical research studies span a variety of clinical areas, such as brain tumor, brain lesions or breast cancer.

Human activity recognition

Over the recent years, human activity recognition is attracting an emerging interest in the area of research. Motion detection applications use a plurality of sensors to appreciate human condition and process the results to create tools for the diagnosis of possible diseases, optimization of physical activity, remote monitoring of vulnerable groups and early intervention in case of emergency. For instance, patients with obesity, diabetes or heart disease, are often required to fulfill a program of activity which follows a training schedule that is integrated within their daily activities. Therefore, the detection of activities such as walking or running becomes quite useful to provide valuable information to the caregiver about the patient’s behavior. MDAKM lab research studies frameworks for automatic recognition of human activities via highly integrated, non-intrusive inertial sensors. The techniques applied for activity recognition include state-of-the-art preprocessing, and data mining methods.

Multidimensional time series analysis

Advances in data collection and data storage technologies led to accumulation of complex multidimensional time series data from many scientific domains such as medicine, biology, economics, geology etc. A well known example of multidimensional time series data from the medical domain is EEG (electroencephalogram) in which many signals derived from different brain regions co-evolve in time. A challenging research area, in which our lab focuses and contributes with innovative solutions, is the knowledge discovery from such complex multivariate temporal datasets. Knowledge discovery includes event detection (such as microstructure events during sleep, K-complexes, or epileptiform discharges), similarity indexing and analysis, high-dimensional pattern classification and unsupervised clustering. Multilinear algebra, and more specifically tensor analysis, offer tools that can simultaneously inspect the underlying information of higher order biomedical data. The nature of this type of data is known to span across multiple dimensions such as space, time, frequency, function, or scale. Tensor analysis of time series finds practice in classification, correlation analysis and clustering schemes.


Our work in bioinformatics is centered around two main axis: analysis of gene expression maps and analysis of protein molecular surfaces. Voxelation is a relatively new method for obtaining gene expression patterns in the brain. It employs high-throughput analysis of spatially registered voxels to produce 3-D maps of gene expression.  Gene expression in the mammalian brain holds the key for understanding neural development and neurological disease. Our lab has done extensive work on large scale data analysis for identifying the relation between gene expression maps obtained by voxelation and gene functions. Moreover, in the areas of drug design, pharmacology and 3D molecular modelling and in order to speed up the drug discovery process considering the rapid increase in genomic and structural database sizes we are developing novel strategies to perform effective and efficient similarity searches and molecular docking experiments using protein molecular surfaces.