RESULTS FROM A NEURAL NETWORK TRIGGER FOR BEAUTY SEARCH AT THE SPS
C. Baldanza, F. Bisi, A. Cotta-Ramusino, I. D'Antone,
L. Malferrari, P. Mazzanti, F. Odorici, R. Odorico, M. Zuffa
INFN ANNETTHE/Bologna
C. Bruschini, P. Musico, P. Novelli
INFN/Genoa
M. Passaseo
CERN/Geneva
ABSTRACT
Results from a neural trigger hosted by experiment WA92, beauty particle
production from 350 GeV negative pions on a fixed Cu target, are presented.
The neural trigger has been used to send on a special data stream events to
be analyzed with high priority, adding to events with electrons and muons
also events selected by a non-leptonic signature exploiting microvertex
detector information. The neural trigger module consists of a VME crate
hosting two MA16 digital neural chips from Siemens and two ETANN analog
neural chips from Intel. The crate also includes boards for the
preprocessing of input variables, for interfacing to the WA92 apparatus and
for monitoring by a PC. The four neural chips run independently, for cross-
checks and to show relative performance and stability. During the
experimental run, only the ETANN chips were operational, while the MA16
were still waiting for completion of the debugging of the chip microcode.
Input patterns of up to 32 variables can be processed, although 16
variables were found to be enough for the problem at hand. They are
preprocessed in the neural trigger module from angles and impact parameters
of tracks determined by the WA92 Beauty Contiguity Processor. Training of
the feed-forward nets was done on a workstation. A final tuning for ETANN
with chip-in-a-loop training was unnecessary, since a single layer net
(i.e. a Fisher discriminant) turned out to be adequate and not outperformed
by a two layer net. Event samples used for training derive from the WA92
1993 production run and consists of events already accepted by the standard
non-leptonic trigger. Using the event reconstruction program, they have
been divided in two classes: one containing events with a secondary C3
vertex (three tracks with sum of charges equal to +1 or -1), and the other
containing the remaining events. Certification of the neural trigger
selections was similarly done on independent events from the 1993 run. The
neural trigger operated for two continuous weeks in the 1993 run. For an
acceptance of 15% for C3 events the neural trigger yields an enrichment of
6.6-7.1 (depending on the test sample), which when multiplied by that already
provided by the standard non-leptonic trigger leads to a total C3 enrichment
factor of about 150 (with a C3 acceptance of 1%). The response time of the
neural trigger module is 5.8 microseconds.
Roberto Odorico
Dip. di Fisica & INFN
Via Irnerio 46
I-40126 Bologna
e-mail: odorico@bo.infn.it