41st Study Sessions on Mathematical Modeling in Biology and Related Topics

[Shunsuke Teraguchi]

Dear all,

We are organizing monthly seminar series called "Study Sessions on
Mathematical Modeling in Biology and Related Topics" in IFReC, Osaka
university. Though usually this seminar series are held in Japanese,
this time (41st, July 16th), we will have a talk in English and you
might be interested.
http://sysimg.ifrec.osaka-u.ac.jp/sysbio/
So let me announce it here:

Title：41st Study Sessions on Mathematical Modeling in Biology and Related Topics
"Some elements of One Class Learning and Neurocomputation"
Time: July 16th (Thu), from 10:00 AM
Place: Meeting Room 1 on the 2nd floor of the IFReC Building, Suita
Campus of Osaka University.
Language: English
Abstract: Most machine learnng techniques involve two class learning wherein
one assumes that one has representative data for both "Class A" and
"Class B" and one trains a classifier of one sort or the other on
this data so that on new data the classifier can decide which of the
two classes the new data belongs to.

While in heavy use, for many applications where one is trying to
decide if the data point is in class A or not (e.g. for a disease) it
is rather inappropriate in principle since while one may have
relatively representative data for Class A, it is unusual to have
representative data for class "not A".

A seemingly better goal would be to learn a filter for A or not A just
from data of examples of class A. This is called "one class
learning" and may be the most appropriate setting for some biological
classification tasks.

However, one class results seem to be much more difficult than two
class and work here still is experimental. In this talk we explain
the elements of this paradigm and show how the methodology of
neurocomputation can potentially be used to give good results in some
case especially when combined with certain kinds of aggressive feature
detection.
Our examples involve both "text classification" and fMRI visual
cognitive task classification.

We will touch focus on the neurocomputation paradigm, but also touch
on the support vector machine technology in this setting. We will
also explain how a method like "Genetic Algorithm" can be used for the
feature selection; and try to make clear what are the difficulties in
all of these approaches. If time permits we will touch on some
preliminary experiments in "deep learning" in this area,

Our plan is to explain all of these techniques without assuming
background in any of them.

Neurocomputation Laboratory Site: http://neurocomputation.wordpress.com

Best,
Shunsuke Teraguchi
--
----------------------------------------------------------
Shunsuke Teraguchi
Quantitative Immunology Research Unit
Specially Appointed Assistant Professor
IFReC, Osaka University
3-1 Yamada-oka, Suita, Osaka 565-0871, Japan
E-mail: tera...@ifrec.osaka-u.ac.jp
----------------------------------------------------------