ClusterID:Tutorials
Describe the JAS environment and the event loop/event processor framework. Describe the use of a neural net.
We include a series of tutorials illustrating many of the uses one may wish to make of ClusterID. To do the tutorials one must already have some proficiency using Java Analysis Studio (JAS) for Linear Collider Detector (LCD) studies. JAS is a Java language based analysis system and LCD is an extension of JAS which is a system for reconstruction and analysis of data from a Linear Collider detector. If one needs to become familiar with JAS and/or LCD then go to the sites linked.
For those already familiar with JAS/LCD, it may be helpful to review the event loop structure.
ClusterID uses neural nets and it maybe helpful to give a brief discussion of how they work. The particular neural net used is called, cjnn, and was written as an extension to JAS. A neural net is a tool for calculating weighting factors similar to probabilities for various hypotheses based on input information. In ClusterID the hypotheses are of the form, "This calorimeter cluster of hits was created by a particle of type X", where X might be photon or neutral hadron or... The input data is a set of numbers, discriminants, that measure properties, discriminators, of the cluster. For example, the length of the cluster. In the processing of an event a cluster builder is called as a stage in the event loop. The cluster builder loops over all the calorimeter hits and groups them into clusters based on some criteria. Then the next stage processor would be ClusterID which would loop over all the clusters in the event, calculate all the discriminants for a cluster and pass the set of discriminants for the cluster to the neural net to determine weights for each possible hypothesis it knows about.
Tutorials:
UsingANetForPerformanceEvaluation:
MakingANet:
EvaluatingCalDesigns:
EvaluatingDiscriminators:
ImprovingResolution:
AnalyzingClusterBuilders:
AnalyzinigClusterIDAlgorithms:
DiagnosticGeneratorStudies: