main page:
|
| Sicco
Pier van Gosliga |
 |
Information:
I am a PhD student at at the
Intelligent Systems Lab
Amsterdam (ISLA)
of the Informatics Institute, University of Amsterdam (UvA).
Furthermore, I am a full time employee at Netherlands Organisation of
Applied Scientific research (TNO),
where I work on probabilistic models
for a wide range of challenging problems.
At Both TNO an the UvA, my main research focus is on using
probabilistic
models in a distributed and decentralised setting. This means reasoning
on
belief networks that are not present at centralised all-knowing
location, but rather the product of many networks at different
locations in a dynamic environment. Mostly in the context of
data
and sensor fusion systems. Typical applications include: situational
awareness for traffic and crisis management.
Software:
I am currently working on an inference engine
for belief
updating in distributed belief networks. In the meantime, please have fun with this Java Applet which uses
a localized version of this engine. It can be used to embed a belief
network on your website, like this:
For more information visit the BBN Viewer Applet page.
My
research:
The following papers that may give you a good impression of my research
interest.
- B.W. Wisse, S.P. van
Gosliga, N.P. van Elst, A.I. Barros,
"Relieving the elicitation burden of Bayesian Belief Networks", 6th Bayesian Modeling
Applications Workshop on UAI 2008, Helsinki, Finland, July 2008 (
 pdf)
This article addresses a common problem that we encounter at TNO when
modelling belief networks. We frequently use child nodes with multiple
parents. In that case the CPTs grow to an inpracticle size. We
wanted to have a usable way to elicit just enough priors and
information to fill the whole CPT, which led to the development of
EBBN. Remember: real world models are usually not restricted to
booleans.
- S.P.
van Gosliga, I. van de Voorde,
"Hypothesis Management Framework: a flexible design pattern for belief
networks in decision support systems", 6th Bayesian Modelling
Applications Workshop on UAI 2008, Helsinki, Finland, July 2008 (pdf)
Another
problem that we encountered at TNO involved the design of
belief networks when causilty is not known or a topic of debate between
domain experts. We suggest HMF as a pragmatic and workable soltution to
overcome this difficulty. Visit the HMF page for more information.
- S.P.
van Gosliga, R.T. van Katwijk, P. van Koningsbruggen, "Real-time
traffic monitoring with Bayesian belief networks", 11th World Congress on
Intelligent Transport Systems (ITS-2005), San Fransisco,
USA, November 2005
(pdf)
Traffic
management
has been one of the application domains that I have been working on in
the past. It has proven to be a very interesting application domain:
its is dynamic, noisy and does easily lead to large and complex models
that combine both expert opinions and sensor data. However,
my
PhD project has been given the priority lately.
- S.P. van Gosliga, M.G.
Maris, "Communication
cost in Distributed Bayesian Belief Networks", 16th Belgium-Dutch
Conference on Artificial Intelligence (BNAIC-2005), Brussels,
Belgium, October 2005
(pdf)
I
am specifically
interested in distributed inference. Therefore I started exploring the
communication cost of belief updating at the start of my PhD project.
This paper compares lambda pi message passing
and the cost of belief updating via cliques. Note: this paper does not
address the communication cost of compiling a junction tree.
Contact information:
Intelligent Systems Lab Amsterdam
Informatics Institute
University of Amsterdam (map)
Kruislaan 403
1098 SJ Amsterdam
The Netherlands
mobile: +31 (0)623324096
phone: +31 (0)205257517
room: F1.16
e-mail: s.p.vangosliga@uva.nl
website: www.science.uva.nl/~spg
(usually
on Monday and Tuesdag) |
TNO Defence,
Security and Safety
Information and Operations
Oude Waalsdorperweg 63 (map)
Postbus 96864
2509 JG Den Haag
The Netherlands
mobile: +31 (0)623324096
phone: +31 (0)703740230
room: 3M14
e-mail: sicco_pier.vangosliga@tno.nl
(Wednesday
until Friday) |
|
