PALESTRA PROF. FABIO RAMOS DATA: 23  de novembro de 2012 LOCAL: sala  401L.
Prédio Cardeal Leme, PUC-Rio HORÁRIO: 10 às 12h

Title: Machine learning for data fusion and the Big Data question

Abstract:  Recent   years  have   seen   an   enormous  proliferation   and
availability  of data typically collected from sensors embedded in computer
devices. Most  people have mobile  phones with  an integrated  camera,  GPS
and accelerometers. Mobile  robots can possess  radars, 3D laser  scanners,
and  hyperspectral  cameras.   Larger -scale  examples  include  the mining
industry  where  several  sources  of  information  (core  drilling,   cone
logging,  spectrometry)  are  available  to  predict  the ore concentration
and assess the quality of the product. However, immense quantities of  data
are not  necessarily  useful unless we   develop methods to  interpret  and
represent  multi-modal   information efficiently.   In  this  talk  I  will
present methods to  jointly infer multiple  quantities from various  sensor
modalities,  at  different space   and time   resolutions. As   an example,
consider the problem  of estimating a  real-time spatial-temporal model  of
pollution  dispersion  in  a  river  using  mobile  platforms.   Given  the
technology  available, the vehicle   can sense biomass,   temperature,   PH
and   many    other   chemical/physical    quantities.  Understanding   the
relationships  between   these  quantities   can significantly  improve the
accuracy  of   the  method   while  reducing   the  uncertainty   about the
phenomenon. I  will show  a set  of techniques  for nonparametric  Bayesian
modelling  that   address the   challenges  in  spatial-temporal  modelling
with heterogenous sensors.   In particular:  1)  how  to define   exact and
sparse models  that are scalable to  large datasets;  2) how  to  integrate
data  collected   at  different  support and  resolutions;  and  3)  how to
automatically learn relationships  between different  quantities   in  real
-time,   from  mobile   platforms.  I   will   show applications  of  these
methods to   a number   of problems   in robotics,   mining exploration and
environmental monitoring.

Short   bio:  Fabio  Ramos  is   a Senior   Lecturer  at  the  School   of
Information Technologies, University of Sydney, and an ARC Discovery  Early
Career  Fellow.  He received   the   B.Sc.  and   the   M.Sc.  degrees   in
Mechatronics  Engineering at University of Sao Paulo,  Brazil, in 2001  and
2003  respectively,  and  the   Ph.D.  degree  at  University  of   Sydney,
Australia, in 2007. From 2007 to 2010 he was  an ARC research fellow at the
Australian   Centre  for  Field  Robotics (ACFR).  He   has  over  80  peer
-reviewed  publications   and  received   the  Best   Paper  Award   at the
International Conference  on Intelligent   Robots and  Systems (IROS)   and
at the Australian Conference  on  Robotics  and Automation  (ACRA). He   is
an  associate  editor for  ICRA, IROS,  RSS and  a program committee member
AAAI, IJCAI and  IPSN. His research focuses on statistical machine learning
for large-scale data fusion problems with applications to robotics, mining,
environmental monitoring and healthcare.