|The goal is to study human inner and outer behavior in order to find possible brain correlates with the (outer) expressive behavior. In the context of behavioral diseases (e.g., autism, schizophrenia, Alzhaimer, Mild Cognitive Impairment, etc.), the main idea is to exploit computer vision and pattern recognition techniques to analyse nonverbal human behavior (face, posture, gesture, etc.) as well as neuroimaging data so to identify possible correlations or characteristic biomarkers. This research not only would support early diagnosis of the pathology but also the monitoring of the effects of the pharmacological treatment.
|Biomedical data processing
|Feature extraction and integration from multi-modal data using multi-scale sparse representations for the definition of numerical biomarkers. Pattern recognition and machine learning for medical imaging and behavioral analysis.
|Neuroimaging Data Analysis
|This domain regards the analysis of data coming from sensing devices measuring the brain strucural and functional information. The main utilised device is Magnetic Resonance Imaging (MRI) in its various modalities such as diffusion, structural, and functional MRI, as well as other sensors like EEG, fNIRS, MEG. The main goal is to better understanding brain functions by means of an integrated functional and structural analysis of the brain connectivity or of specific brain regions. This investigation is mainly performed with reference to neurological disorders - like autism and schizophrenia - and in comparison with control (healthy) subjects.
|Decision Tree Optimization
|One of the most studied data mining tasks in the literature is the classification task, consisting of learning a predictive relationship between input values and a desired output. A classification problem can also be viewed as an optimization problem, namely as the problem of building a model that maximizes the predictive accuracy—the number of correct predictions—in the test data (unseen during training). We are interested in the problem of optimizing the construction of decision trees. Decision trees are widely used in data mining and machine learning as comprehensible representation models, given that they can be easily represented in a graphical form and also as a set of classification rules, which can be expressed in natural language in the form of IF-THEN rules.
|The main focus is on the study and development of automatic techniques and models able to extract information from real world data, typically in terms of classes or clusters. Special attention is on probabilistic models - like Hidden Markov Models, Mixtures, Topic Models - and on kernel machines - like Support Vector Machines. In these contexts the interest is in designing novel models/methodologies, like hybrid generative-discriminative methods, generative embeddings and kernels, novel classification or clustering schemes, model selection techniques and others. The focus is on reasoning on representation issues (how to extract features, how to process the original problem space) as well as on unconventional employment of standard techniques (like boosting or SVM for clustering). Another field of interest is the processing of sequential data (using for example Hidden Markov Model).
******** CSS e script comuni siti DOL - frase 9957 ********p>