Measurement Techniques in
High Energy Physics at KTH
Announcing the
Virtual
Intelligence - Dynamic Neural Networks VI-DYNN'98 Workshop
A workshop on PCNN theory and applications,
as well as dynamic neural network models
Tobias, Maxim, Thomas, Clark
Personnel:
At KTH/Physics Dept. - Frescati, Stockholm
Sölve, Åge,Clark,Givi,Guy,Kenneth
-
Åge Eide:
assoc. professor, Ostfold College, Halden, Norway. Frequent Collaborator
-
Givi Sekhniaidze:
Visiting researcher from Georgian Academy of Science.
-
Guy Paillet: Neuroptics, Montpellier, France.Occasional Collaborator
-
Kenneth Agehed:
research engineer.
-
Geza Szèkely:
Debrecen, Hungary.Frequent Collaborator
Jason
Mary Lou and Bruce
-
Bruce Denby: University
of Versailles, Paris, France Occasional Collaborator
-
Jason M. Kinser:Jason
Kinser, Research Associate Prof., George Mason Univ., Manassas, Va., USA
-
Mary Lou Padgett:
Auburn University, Auburn, Alabama, USA.Occasional Collaborator
ANNSR2: Applications of Neural Networks in Scientific
Research
The ANNSR2 Collaboration during a meeting in Stockholm.
Including our work in Pulse Coupled
Neural Networks
Don't miss our
Overview
Measurement techniques and data processing applied to the micro world
often involves a whole different approach than that for our ordinary macroscopic
world. This is for a number of reasons: one must usually destroy the very
object that one wishes to study, one must describe the object of interest
from examining the debris of that destruction, the background noise can
be 100 million times stronger that the desired signal, the data volumes
generated can consume thousands of high density magnetic tapes, etc.
The data processing chain includes detectors (e.g. particle trackers
like wire chambers), fast electronics for event filtering (or "triggering")
so that only the small number of events of interest are actually written
to tape, highband width links to computer clusters for more event filtering
before writing to tape. Further data processing then occurs off-line as
various "cuts" are applied to the data to enhance the signals. We study
techniques that cover each of these links in the chain from the data acqusition
to on-line and off-line filtering and, finally, to analysis of large databases.
For further discussion, see the following:
Diploma Work
If you are interested in doing your diploma work with us, click here
in order to find out what projects are available right now.
Neural Networks
Clark Lindsey and Givi Sekhniaidze holding IBM ZISC neural
network PC and VME cards respectively.
The primary emphasis of our group is fast pattern recognition and
signal processing, especially with the use of neural networks implemented
in hardware. Such hardware systems could for example, be used in the real-time
triggering systems of high energy physics detectors. Neural networks allow
for robust pattern recognition for data with large statistical fluctuations
and noise. Their inherent parallel architecture provides fast performance
if implemented in hardware (as opposed to the usual simulations in conventional
sequential cpu's.) Our group has built prototype or purchase neural networks
using such devices as the Intel ETANN, IBM ZISC036, and the Bellcore CLNN.
Applications studied include pion recognition from Cherenkov detectors,
and secondary vertex event tagging using silicon vertex detector tracking
signals. See our Neural
Networks in High Energy Physics page that we have created (along with
Prof. Bruce Denby of Univ. of Versailles) for further information. Also,
see the hypertext version of our review
paper on hardware neural networks (see also the new framed
version for Netscape 2.0 users).
Some of the current applications we are studying include:
-
Neural networks for small satellite Star Trackers and satellite control.
-
Beta testing of the TOTEM neural network chip in a PC card. This chip
was built by a group at Univ. of Trento and uses a novel "Restricted Tabu
Search" for fast on-chip learning. It has 32 neurons and can forward process
a 16-16-1 network in 2microsec.
-
Data Mining
with neural networks. That is, use a neural network to look for generally
related topics or concepts rather than exact matches. Recently, we obtained
a board from Microway.MRT, Norway, which can search a 160Mbyte memory in
1 second for a given 32 byte character string. We are investigating how
to combine this capability with neural networks.
-
Combining wavelet preprocessing of inputs to neural networks. A VME
card with 2 Aware Corp. wavelet chips has been built.
-
Use wavelets to eliminate spikes in cellular phone noise.
There have been several neural network and wavelet hardware cards constructed
(*) or purchased by the group over the past few years. These include:
ATM
We are currently also engaged in the NEBULAS project (also called RD-31).
This project is studying the use of the Asynchronous Transfer Mode ( ATM),
often referred to under the title of broadband digital networks, for use
in data acqusition in future experiments at such accelerators as the LHC
at the CERN laboratory in Geneva,
Switzerland. Collaborators in NEBULAS include groups from CERN, Saclay,
Uppsala University, and the Alcatel company. ATM was developed by the telecommunications
industry but has now become of great interest for general networking, including
LAN's of PC's and workstations. The base rate is 155Mbits/sec (compared
to 10Mbits/sec for ethernet). Using fixed length packets of 53 bytes (5
byte header + 48 byte payload), the packets travel "point to point" over
dedicated links, rather than via shared media as in ethernet or token rings.
Switches read the packet headers to determine the destination address of
the packets. ATM treats data, audio and video equally (higher layer protocols
distinguish the packets) and so will become a central part of the "multimedia
superhighway" of the future. NEBULAS hopes to take advantage of the rapid
developments of this technology to overcome many of the limitiations of
current data acquisition systems that have serious bottlenecks, e.g. data
buses, that will hamper the huge data rates at LHC detectors. See the High
Speed Interconnect page at CERN for more information on ATM and other
broadband technologies. Also, see the list of ATM
links. We have recently tested two Olicom
155Mbit/sec EISA cards and used them to send ATM packets between two
PC's.
Part of the NEBULAS-Collaboration during a meeting in Stockholm.
KTH-Frescati Home Page
Curator:Clark
S. Lindsey -lindsey@particle.kth.se)
