Dr Giorgio Roffo
- Research Associate (Computing Science)
telephone: +44 141 330 1651
My research focuses on building new approaches to enable machines to analyze, recognize, and predict human communicative behaviors during social interactions. This multi-disciplinary research topic overlaps the fields of computer vision, machine learning, and social psychology.
My current research focuses on bringing deep learning solutions to tease out the structure of the elaborate code behind social interactions (Human-Human & Human-Robot), making it possible for machines to read and write human body language.
What makes this research area different from classical computer vision is the needs to deal with multiple communities and modalities (e.g., Visual, Auditory, Haptics, Verbal, etc.). This is why we talk about multimodal machine learning.
Currently, we are working on multimodal deep learning and Feature Selection, where different specialized deep networks are trained in a natural way for each data source. As a result, deep networks allow us to retain a homogeneous interface where all the features are mapped in the same feature space. Novel strategies can be developed to accurately sense and interpret human social signals and social context with the help of feature selection techniques embedded in deep networks. In our opinion, next-generation computing needs to include the essence of social intelligence to become more effective and possibly understanding a facet of our communication better than we do ourselves.
Personal website: http://giorgioroffo.uk
nVIDIA GPU Grant 2017. NVIDIA GPU grants are intended to enable researchers to begin a new project and/or gain the preliminary results to support a larger proposal to other funding agencies (see GPU Grant Program).
- Feature Selection Library (FSLib 2018) is a widely applicable MATLAB library for feature selection (attribute or variable selection), capable of reducing the problem of high dimensionality to maximize the accuracy of data models, performance of automatic decision rules as well as to reduce data acquisition cost (FSLib was awarded by MATLAB in 2017).
- Visual Object Tracking (VOT2016) Trackers repository. Each tracker is referred to as NAME (Appendix Nr NN), where NAME is the name given to the tracker and NN is the tracker number referred on the VOT paper. Our contribution is DFST (Appendix Nr 39)