Research

  • Main expertise and experience in computer vision and pattern recognition, machine/deep learning, image and signal processing, and neuroimaging (data analysis).

  • Main research focus lies on deep learning, specifically on domain adaptation and generalization, multimodal deep learning, learning with privileged information, zero/one/few-shot learning and, in general, learning with “imperfect” data (i.e., unlabelled or partially labelled, few, annotated with noisy or weakly labels, class imbalanced, biased, or a combination of these). I am recently fascinated by approaches leveraging generated synthetic data in order to reach same performance as using real data (so getting rid of costly real data collections and annotations). Related applications involve classification and recognition, in supervised and unsupervised scenarios, including (fine-grained) activity recognition.

  • Particular interest in multimodal social signal processing approaches for the analysis of (even nonverbal) human behavior, with main applications related to surveillance and security, human-human and human-machine interaction, ambient intelligence, and retailing. Also major experience in typical industrial application domains such as visual inspection and automation.

  • Concerning the biomedical area, the main work and interest go to neuroimaging data analysis, namely Magnetic Resonance Imaging and, in particular, in the study of neural correlates responsible of (social) behavior, aiming at the investigation of behavioral neurological pathologies (e.g., schizophrenia, autism, etc.) and brain function understanding in general. We deal with these problems adopting a “connectomics” approach, mainly exploiting and integrating structural and functional information.

  • “Recent” interest in the use of neuroimaging data in complex visual recognition tasks. In short, the driving idea lies in exploiting neuroimaging data (electroencephalography, EEG, at first instance) and visual data to improve the performance of deep learning models. The goal is to assess whether brain recordings (e.g., EEG) would be useful to tackle difficult recognition tasks, either when processed together with visual data, or when used as privileged information, available only at training but not at test time.

  • Former experience in underwater vision (acoustical and optical), data fusion and sensory integration with applications on (underwater) object detection and recognition, object and scene reconstruction.

Research keywords

Computer Vision

Machine Learning

Deep Learning

Multimedia

Pattern Recognition

Underwater Vision

Signal, Image and Video Processing

Pubs repositories and CV

CV and main publications (pdf), updated at November 2023