1. Multidimensional classification of Kleine Welle triggers using LIGO data.
Soma Mukherjee for LIGO Science Collaboration
Center for Gravitational Wave Astronomy
University of Texas at Brownsville.
GWDAW 10, Dec Brownsville, TX, USA.

2. Multidimensional visualization and classification
Multidimensional visualization is a technique of visualizing
proximity data i.e. data where objects are characterized by
dissimilarity values for all pairs of objects. Maps of these
objects are constructed in Rk by interpreting these
dissimilarities as distances.
Hierarchical classification is a way of grouping data
simultaneously over a variety of scales by creating a cluster
tree. Distances are computed between every pair of objects
and merged together depending on the proximity of these
objects to each other. Objects are first grouped into binary
clusters which in turn are grouped into larger clusters until
a hierarchical tree is formed.

3. Why MDiC in trigger analysis ?
MDiC enables viewing structures present in the data in a
multiparameter space. Thus, a set of k n-dimensional data vector is
represented as k points in an n-dimensional space.
Often complex data do not reveal embedded structures in simple 1 D
or 2 D plots.
MDiC allows identification of substructures in the data sharing
common properties that can lead to potential tracking of sources and
hence finding vetoes.
MdiC is also the first step towards construction of a training set in
the automated classification pipeline which is the ultimate goal of this
work.

4. Motivation To look for existence of structures or ‘classes’ in the
burst triggers from all channels embedded in
multiparameter space.
To compare the classes thus obtained from several
channels to look for possible correlations.
Automation of the ‘class recognition’ process by using
learning algorithms leading to …
Vetoes.

5. What is being done in the present study ?
Kleine Welle parameters :
Trigger duration
Central scale
Derived parameters :
SNR [ (Etotal – npix) –1/2 ]
Shape information :
10 highest wavelet coefficients.
Optional :
KW central time
Trigger significance
Examples generated using LIGO
S4 data from Hanford 4k
interferometer.
Channel used :
H0:PEM-BSC2_ACCX
GPS : s

6. classification scheme
Analysis pipeline :
Raw time series
Generation of KW trigger database
Gravitational wave channels
Several environmental channels
KW triggers
with duration,
Central scale,
SNR (a cut may
Be applied)
Extract time
series (1 sec
Each way around
the trigger )
from raw frames
Band pass
Compute 10
highest wavelet
coefficients
Hierarchical
classification scheme
Identification
Multidimensional
visualization
Inter-channel comparison
Automated pattern recognition

14. A 1 D projection plot of the SNR that shows how the triggers are
placed in the SNR distribution.

15. On-going … Started analyzing S5 data.
All channels for which KW triggers are available.
Looking into more information on the trigger shape.
Web page under construction where daily/weekly analysis
reports are stored.
Working towards automated trigger recognition pipeline.