Optics Meeting Jan 17 2023 230PM ET
From Moller Wiki
- Classifier for optics: Ciprian slides
Vassu, Kate, David, Andrew, Ciprian, Bill, Zuhal
- (Ciprian) Our present deconvolution algorithm is devised to extract the Moller asymmetry in the ring 5 "open" tiles, by definition - so our kinematic factor should aim for the factor for this asymmetry. The relative rates ("yields") of various processes in each ring/tile type, which are required for this deconvolution, can be extracted from simulation. However, Cip outlined a method to use machine learning techniques (i.e. random forest classifier, boosted decision trees) to extract these yields from tracking data directly. One can used GEM variables (r,r',phi,phi') directly without need to get to the target variables (E', theta phi, Z), and would naturally include any imperfections in the B field, collimator, etc.). Can use a python toolkit scit-learn.org.
- Need to train on simulated data with the appropriate relative rates of all processes.
- Could study effect of B-field imperfections (including simple scaling of mean field), collimator and detector imperfections, with changed training sets.
- Provides important cross-check of yields from the simulation.
- Statistical Error analysis automatically included.
- Can check using sieve data as separate training sets (since these have nicely defined kinematics).
- Separate note: need to study effect of B-field imperfections, collimator and detector imperfections, beam position/raster effects on our polynomial parameterization of optics matrix elements.
- (David) Meeting day/time unlikely to change for this semester. Will know for sure later this week.
- (David) Have asked Andrew to look into creating a Hyperon generator for remoll.
- (Kate) Will contact JLab's new GEANT4 expert (Makato Asai) to see if can get input on problems with the low-energy EM libraries.
Meeting ID: 972 5975 5403