Description: Learning machines are
currently applied in a variety of problems and settings,
including big-size problems (high cardinality and/or
dimensionality), in connection with information processing
methods inspired to the computation processes going on in
living and natural systems, such as Neural Networks,
Evolutionary Computation and Fuzzy Logic based systems.
New theoretical advancement in this field can boost the
methodologies we apply in our applicative projects,
including road traffic forecasting, bioinformatics data
analysis, supportive technology for “fragile” people, and
hedonic IT systems design. In particular, data clustering
is a set of methods that have been studied for a long
time, and yet still offer room for improvement on both the
theoretical and methodological sides. The development of
more powerful and flexible data clustering models offers
the potential for boosting the data mining capability of
automatic systems. During the last decade the development
of relational (spectral and kernel) clustering methods has
opened new perspectives. Possible outcomes of this
research can be the development of effective relational
methods to cope with issues related to large data size,
throughput, and information content, for example by
introducing well-characterized approximations to deal with
high-cardinality data, effective data representations to
tackle high-dimensional problems, or online versions of
relational methods both for low-storage learning and for
learning in the presence of concept drift.
Link to the group or personal
Webpage:
http://www.disi.unige.it/person/MasulliF/ricerca/index.html
Reference Filippone, M., Camastra, F., Masulli, F., &
Rovetta, S. (2008). A survey of kernel and spectral methods
for clustering. Pattern recognition, 41(1), 176-190.