Speaker: Prof. Saman Halgamuge
Time: July 18, 13:00
Venue: 3-528, SEIEE
Abstract:
Unsupervised
learning is used for analysing and clustering data when the expected cluster or
class labels are not available, i.e., we are not aware of the type of
information to be found. When we know only a little about the data labels, it
is still challenging to make conclusions out of the data, although this may be
the case for many real world data mining problems. We name the type of learning
algorithms useful in this scenario as Near Unsupervised Learning (NUL). My
group has been developing NUL algorithms based on the previously proposed
Growing Self Organising Maps. The concept and the algorithm development in NUL
and the application in various biological data mining problems will be discussed.
Some “unusual” features and signatures captured by my team will also be
presented. Real problems attempted using NUL includes the following: 1)
Metagenomics involves the challenging problem of clustering and eventually
labelling genomic data of microbial species that cannot be easily grown in
laboratories. We only know about 2% of these species found on Earth. Could this
be the life form we expect to find on another planet? How do we understand and
use some unique characteristics of microbes living in our environment and our
body? 2) Analysing metabolite profiles of various wheat plants to understand
how some type of the plants can survive droughts is an area where NUL can
provide good solutions, 3) Can we analyse signals coming from biological neural
networks grown on wet labs to differentiate the sick brain tissues from healthy
ones? Our collaborating researchers create mice with genetics based brain
diseases and analyse the brain tissues with and without the introduction of
drugs. Which drugs (for example drugs preventing epileptic attacks) are more
effective on a particular type of sickness? The following research papers summarise
the application of these methods:
[1] C.D. Wijetunge, Z. Li, I. Saeed, J. Bowne,
A.L. Hsu, U. Roessner, A. Bacic and S.K. Halgamuge, "Exploratory Analysis
of High-Throughput Metabolomic Data", Metabolomics, 2013, 9 (6),
1311-1320, Springer. [2] I. Saeed, S.L.
Tang and S. K. Halgamuge, “Unsupervised discovery of microbial population
structure within metagenomes using nucleotide base composition'', Nucleic Acids
Research, Volume 40, Issue 5, 2012, Oxford University Press.
[3] I. Saeed and S. K. Halgamuge, “The
oligonucleotide frequency derived error gradient and its application to the
binning of metagenome fragments '', BMC Genomics, 10(Suppl 3): S10, 2009
[4] A. L. Hsu and S. K. Halgamuge, “Class
structure visualization with semi-supervised growing self-organizing
maps", Neurocomputing, Vol: 71
Issue: 16-18, Pages: 3124-3130, Elsevier, 2008.
[5] C.K. Chan, A.L. Hsu, S.K. Halgamuge and S.L.
Tang, “Binning Sequences Using Very Sparse Labels within A Metagenome'', BMC Bioinformatics, 2008,
9:215, 28 April 2008.
Bio:
Prof. Saman Halgamuge is a Professor of
the Department of Mechanical Engineering and the school wide initiative on
Biomedical Engineering and Associate Dean (International) for the Melbourne School
of Engineering, The University of Melbourne. He completed B.Sc. Engineering
(Electronics and Telecommunications) at the University of Moratuwa, Sri Lanka
and went on to graduate with Dipl.-Ing and PhD degrees in Electrical
Engineering at Technical University of Darmstadt, Germany.
Professor Halgamuge has an outstanding
reputation for fundamental research in Big Data Analytics, Unsupervised
Learning and Bio-inspired Optimization. His research has applications in Bioinformatics,
Mechatronics and Sustainable Energy. He is the co-author of over 250 research papers including 6 books, 20 book
chapters and 80 journal papers with over 4800 citations and h-factor of 28. He
is listed as one of the most cited (top 1%) scientists in the last 10 years by
ISI Essential Science Indicators. At Melbourne, he has completed supervision of
27 PhD students. He serves on the editorial boards of 6 journals including ACTA
journal on Control and Intelligent systems and BMC Bioinformatics. He chaired
12 conferences and served as a member of about 75 conference program
committees. His profile is at http://scholar.google.com.au/citations?sortby=pubdate&hl=en&user=9cafqywAAAAJ&view_op=list_works