About myself:
Hello, My name is Yingxin Wu ( Ying Xin Wu ), English
name: Christine Wu. I come from Guangzhou,
P.R.China.
Currently, I am a PHD student in the School of Information Technologies, the University of Sydney,
My PHD research project is visualizing multivariate networks. Currently I am
using GeoSOM which is a spherical
Self-Organizing Map. I am learning differential geometry, topology and information
geometry.
Email: chwu at it.usyd.edu.au
Please have a
look at my resume. If you have any job to
offer, please let me know. Thank you!
¡¡
Publications:
Thesis:
Ph.D. thesis: Hybrid
Multivariate Network Visualization Combining Dimensional Projection and Graph
Drawing submitted for review.
About
GeoSOM and Multivariate Network:
Visualizing
Multivariate Networks: A Hybrid Approach. Yingxin Wu, Masahiro Takatsuka. IEEE Pacific Visualization
Symposium 2008, Kyoto , Japan
Spherical Self-Organizing
Map using Efficient Indexed Geodesic Data Structure Yingxin Wu, Masahiro
Takatsuka. Special Issue of Journal of Neural Network, Vol .19, Issue
6-7, July - August 2006, pp 900-910. Acknowledgement: It is my fault
that I forgot to put the acknowledgement in my paper. Here I sincerely thanks
Le Song and Kathryn Merrick for their invaluable opinions!! 
Visualizing Multivariate Network Using
GeoSOM and Spherical Disk Layout Yingxin Wu, Masahiro
Takatsuka, Richard Webber. IEEE Symposium on Information Visualization
2006(poster),
Visualizing Multivariate Network on the Surface of
a Sphere. Yingxin Wu, Masahiro
Takatsuka. Asia-Pacific Symposium on Information Visualization (APVIS2006),
Geodesic Self-Organizing Map.Yingxin Wu, Masahiro
Takatsuka. Conference on Visualization and Data Analysis 2005 (5669), Part of
IS&T/SPIE's International Symposium on Electronic Imaging 2005 .
¡¡
Geodesic Self-Organizing Map and Its Analysis.
Yingxin Wu, Masahiro Takatsuka. the 28th Australian Computer Science
Conference, Conferences in Research and Practice in Information Technology,
Vol.38, 2005
Fast Spherical Self Organizing Map--Use of Indexed Geodesic Data
Structure. Yingxin Wu, Masahiro Takatsuka. Workshop on Self-Organizing Maps
2005, WSOM05,
¡¡
About Graph Drawing:
Visualization and
Analysis of Email Networks. Xiaoyan Fu, Seokhee Hong, Nikola S. Nikolov,
Xiaobin Shen, Yingxin Wu, Kai Xu. IEEE Asia Pacific Symposium on Visualization
2007 (APVIS2007),
Visualisation and Analysis of Network Motifs. Weidong
Huang, Colin Murray, Xiaobin Shen, Le Song, Yingxin Wu, Lanbo Zheng. IEEE 9th International Conference
Information Visualisation,
Visualisation and Analysis of Large and Complex Scale-free
Networks. Adel Ahmed, Tim Dwyer, Seok-Hee Hong,
Colin Murray, Le Song and Yingxin Wu.
Eurographics / IEEE VGTC Symposium on Visualization.
Wilma Scope Graph Visualization. Adel
Ahmed, Tim Dwyer, Colin Murray, Le Song, Yingxin Wu. Student First Prize, IEEE InfoVis Contest 2004.
¡¡
About
Color Picker:
Three Dimensional Colour Pickers.Yingxin Wu,
Masahiro Takatsuka. Asia-Pacific
Symposium on Information Visualization (APVIS2005),
Multivariate Networks Visualization:
A
multivariate network is defined to be a data set which contains both
multidimensional data and relational data. For example, an international
trading network describes the trading relationship between countries. Trading
for different commodities (e.g. food, machines) or trading at different time
form various relationships between the countries. Beside the relationships, a country
might contain gross domestic product, population, literacy level or military
expenditure as the attributes. Similarly, in a social network, people have
different attributes (age, height, weight, education level) and relationships
(e.g. marriage, friends, colleges etc).
Therefore
the word ¡°multivariate¡± has two meanings:
¡¤
The vertices have multiple attributes.
¡¤
The same set of vertices has different relationships.
A multivariate
network is not a simple addition of multidimensional data and relational data.
Entities¡¯ attributes and relationships may influence each other. For example,
people's background, habits and personality affect the types of friends he/she
has. Companies with good reputation and strong economic background are easier
to establish business relationship with other companies. Visualization of a
multivariate network should show reveal the vertices¡¯ attributes, relationship
and the ``interactions'' of these two aspects.
Our
visualization approach combines multidimensional data projection techniques and
graph drawing algorithms. Vertices
are considered as data points in high dimensional attribute space. We use the
spherical Geodesic Self-Organizing Map (GeoSOM) to determine the initial layout
of a multivariate network. The training process is modified to consider both of
the vertices' attribute similarities and the graph distance distribution:
Vertices that have similar attributes and are close with respect to graph
distances will be mapped onto proximate regions on the GeoSOM. After that,
positions of the vertices are adjusted by a graph drawing algorithm to remove
the vertex-overlaps and reduce edge crossings. This method is called the hybrid approach. An example is shown in
the following figure:
Figure
1. Visualization of a multivariate
network using the hybrid approach. The network contains 48 vertices and 67
edges. Vertices have 10 attributes and can be classified into three clusters
according to attribute values.
In the
hybrid visualization, structure of the high dimensional attribute space (i.e.
clustering of the vertices) is shown by the color background. Colors are
linearly changes from blue, through cyan and yellow, to orange. Blue denotes
the smallest variance between vertices, and orange the largest. Therefore,
vertices mapped to the same blue region of neurons are close in
high-dimensional space and can be considered to be within the same cluster.
Cluster boundaries appear as yellow or red bands surrounding the clusters. For example, in Figure 1, the vertices
are located in three attribute clusters. Relationship between vertices is
represented as edges on top of the GeoSOM. In such way, viewers are able to see
whether attribute similarities have any impact on the vertices¡¯ relationships.
For instance, vertices in Figure1 are more likely to be connected to vertices
within the same attribute cluster than between clusters. In addition, using the
component plane of SOM, the visualization can also display the distribution of
selected attribute across the network, which reveals the correlation between
that attribute and the vertices¡¯ relationship. Pictures of the component plane
can be found in the gallery.
The
GeoSOM is a spherical Self-Organizing Map. Since our flat computer screens
cannot show the front and back part of a 3D object at the same time, it is
difficult for viewers to maintain mental map of the entire GeoSOM. To solve the
problem, we implemented an interface to project GeoSOM onto a 2D plane using
cartography approach. Currently we choose the Wagner III pseudocylindrical
projection to transform the spherical surface onto a 2D plane (see Figure 2).
Using our interface, viewers can easily choose any point on the GeoSOM to be
the center of 2D projection. Changing the orientation or center point of the 2D
map only requires re-projection. However, edges going through the split line of
the sphere are broken into two parts and draw separate at the left and right
end. Figure 1 is an image projected from the spherical GeoSOM. Edge (20,45) is
divided into two edges.
Figure 2
The Spherical GeoSOM is projected into 2D plane using cartography approach.
Orientation The user selects two points on the sphere using a mouse. The first
point A will be the center of the 2D projection. The second point B, together
with A, define a plane going through the sphere's center O. Suppose Vector OC
is the plane's normal, then the sphere's central axis can be obtained by the
cross product of Vector OC and
Vector OA. This axis intersects the sphere at two points, N and S, which become
the ``North'' and ``South'' poles of the sphere. The geodesic arc going through
these two points and opposite to A becomes the split line on which to open the
sphere.
Gallery:
Visualization of
International Trading Networks
For space
reasons, in the following visualizations, country names are displayed in their
standard three letters abbreviation form. The trading values are mapped to
different transparency levels of the edges. The higher the trading value, the
darker of the edges. Furthermore, in order to show the balance of trade, two
ends of each edge are of different widths. The widths are proportional to the
corresponding countries' exporting value. Note that in the visualizations, countries¡¯ positions are grouped by
their similarities in attribute values instead of real geographical positions.
¡¤
Visualizing International Cereals and Manufacture Metal
Trading Networks in 1994 (Please use IE to
open the link instead of Mozilla FireFox)
These
data sets were extracted from the Pajek database. They describe the cereals and
manufactured metal trading between 80 countries in 1994. Each country has four
attributes: GDP per capita, GDP Growth, population and population growth. The
cereal trading network has 230 edges and the manufactured metal network has 248
edges. The edges are directed, pointing form exporting countries to exporting
countries. Some edges are double directed because both countries export the
same types of commodity to each other. Each edge is associated with the
corresponding trading value in US dollars. Due to data availability, each
country in the network only has the aforementioned four attributes . It is very
easy to incorporate more attributes into the visualization once the data are
available.
¡¤
Visualizing International
Military Expenditure and Arms Transfer Networks in 1983-1991 and 1992-1999 (Please use IE to
open the link. Seems some versions of
Mozilla FireFox cannot open the webpage. Please wait a little while for
IE to load the pictures because the pictures are of large size. Thank you!!)
These
data sets are collected from two sources: the Stockholm International Peace
Research Institute (SIPRI) and the USA Department of State. SIPRI maintains an
Arms Transfer Database which records the international trading of major
conventional weapons from 1950 to present. Each record lists the export/import
country and the transferred volume. The USA Department of State generates a
yearly report called the World Military Expenditures and Arms Transfers
(WMEAT). The WMEAT contains lengthy tables listing various indicators which
measure the countries' military expenditures. However, I am only able to
collect reports from 1983 to 1999. From each report, we extract eight
indicators:
1.
Gross national product per capita (GNP per capita)
2.
Military expenditures per member of the armed forces
(ME/AF)
3.
Percentage of military expenditure to gross national
product (ME/GNP)
4.
Percentage of defense cost to the central government
expenditures (ME/CGE)
5.
Military expenditures per capita (ME per capita)
6.
Armed forces per 1000 people
7.
Percentage of military imports to non-military imports
8.
Percentage of military export to non-military exports
.In order
to show the networks' changes over a period of time, the data are divided into
two sub-periods 1983-1991 and 1992-1999 (before and after the cold war). The
country attributes are averaged over the years in each period. Some countries
in war such as
If you
are interested in more details of the pictures shown in the gallery, please
read this document which is the draft
of Chapter 7 in my PHD thesis. Please email me any comments and opinions,
especially whether you find our visualization helps you to understand the
international trading networks. Thank you very much.
Link
to My Friends:
Le Song Zhao ShanHeng Gang
Cheng Juan
Qin Yunjie
Li
¡¡
Useful Links
3D Geometry:
Interesting algorithms about 3D geometry.
¡¡