CS Events Monthly View

While unprecedented amounts of data are now readily available, it is
often not clear how to exploit these forms of data for machine learning,
natural language processing, and artificial intelligence. In this talk,
I will describe a series of results on methods that integrate and learn
from multiple, possibly quite heterogeneous kinds of data. In natural
language processing, for instance, these allow us to produce massively
multilingual word representations, covering over 200 languages. Combined
with data integration methods, we can also produce representations of
entities and large-scale knowledge graphs such as UWN/MENTA, one of the
largest multilingual taxonomies, and Lexvo.org, a major hub in the Web
of Data. Finally, I will outline new methods for mining from multiple
modalities, including images and video, that have the potential to take
us towards cognitive computing.