BEGIN:VCALENDAR VERSION:2.0 PRODID:-//CERN//INDICO//EN BEGIN:VEVENT SUMMARY:CIDeR-ML General Meeting DTSTART:20261002T000000Z DTEND:20261002T003000Z DTSTAMP:20261006T092800Z UID:indico-event-10736@indico.slac.stanford.edu DESCRIPTION:https://u-tokyo-ac-jp.zoom.us/j/83932834349Link to recording\n \n\nQuick recap\nYiyang presented on electron tracking with simulation-bas ed inference and probabilistic programming language in a toy chain contact or model. Yiyang discussed the challenges of reconstructing low-energy ele ctrons in water due to severe multiple scattering\, explaining how traditi onal methods using marginalized PDFs may lose important covariance informa tion among photon tracks. He presented two approaches: simulation-based in ference using deep sets to approximate likelihood ratio corrections\, and probabilistic programming language implementation that translates Gen4 phy sics processes into a probabilistic framework. Both methods found that whi le photon direction information contains measurable non-IID structure beyo nd marginalized profiles\, it does not significantly improve initial elect ron angular resolution due to degeneracy in track reconstructions. The wor k did demonstrate that scattering severity can be measured and used for ev ent classification\, and timing information provides stronger constraints on electron directions when PMT-TDS precision is sufficient. Junjie sugges ted Yiyang present more detailed breakdowns of the methods in future sessi ons to better digest the complex technical content.\nNext steps\nYiyang\n\ nPrepare and present a more detailed presentation on one of the methods (s imulation-based inference or probabilistic programming language) at the ne xt meetings.\n\nSummary\nElectron Tracking Simulation Research\nYiyang pre sented a study on electron tracking with simulation-based inference and pr obabilistic programming language in a toy chain contactor model. The resea rch aimed to improve angular resolution for low-energy electrons in water by leveraging information beyond marginalized electron tracks\, as traditi onal reconstructions lose covariance among tracks. Yiyang demonstrated tha t while simulation-based inference could recover track-aware likelihood an d classify events based on multiple scattering severity\, it did not signi ficantly improve angular resolution itself\, though the log-likelihood rat io could be used to indirectly boost resolution by identifying more track- like\, direct events.\nP0 vs P-TRACK Comparison Discussion\nYiyang explain ed the difference between P0 and P-TRACK\, noting that P0 uses a manual te mplate to multiply marginal PDFs while P-TRACK requires numerical approxim ation through a classifier trained on both ID and Gen4 track samples. Junj ie questioned how the photon propagation data is generalized into a likeli hood distribution\, and Yiyang clarified that a shared photon encoder proc esses variable photon numbers into fixed-dimensional features before class ification. The discussion revealed that their toy model currently only tra cks electron scattering without light transportation\, focusing on electro n scattering at low energies.\nProbabilistic Programming for Geant4 Physic s\nYiyang discussed the challenges of implementing probabilistic programmi ng for Geant4 physics processes\, including variable length dimension infe rence and the use of fixed track segment numbers with varying track length s to improve efficiency. They calibrated the model using Geant4 input as a benchmark and found that while photon directions provided some measurable non-IID structure\, they did not significantly improve initial electron a ngular resolution due to degeneracy. Yiyang concluded that timing informat ion could provide stronger constraints on electron initial direction but r equires precise PMT-TTS\, and suggested future improvements including cros s-energy calibration and using classifiers instead of marginal templates t o train the PPL model.\n\n\n\nhttps://indico.slac.stanford.edu/event/10736 / URL:https://indico.slac.stanford.edu/event/10736/ END:VEVENT END:VCALENDAR