Login

CIDeR-ML General Meeting

→ America/Los_Angeles
Description

https://u-tokyo-ac-jp.zoom.us/j/83932834349

Link to Recording

Quick recap

This meeting was a roundtable discussion focused on updates and progress reports from various team members working on detector calibration and simulation projects. Naomi introduced herself as a PhD student working on the Hyper-Kamiokande far detector and reported on her work using LUCiD to set up calibration scenarios with a light injection system, though she identified concerning amounts of backscatter at the injection site that may be related to a known bug in LUCiD. Riya presented her comparison of light injection simulations between SKDETSIM and LUCiD, showing generally good agreement but some mismatches that could be due to differences in PMT positioning and photon numbers. The team discussed technical details about PMT positioning, photon injection rates, and potential sources of discrepancies between the two simulation tools. Zhenxiong reported on her presentation at the WCTE collaboration meeting about probing angular response using the SIREN model with cosmic muon data, though there was discussion about uncertainty propagation through SIREN parameters and how to connect those to physical parameters. Junjie provided an update on ongoing work investigating limitations of the sirenTV model, particularly focusing on frequency domain analysis and the impact of omega-0 on model performance. The conversation ended with a discussion about the timeline for publishing their paper, with the team agreeing to aim for completion by the end of September, contingent on resolving current performance limitations in the SIREN model.

Next steps

Naomi

  • Check the positions of the light injectors and PMTs in LUCiD to ensure they match those in WCSim and the actual injection locations.
  • Share the event displays from LUCiD in the chat for comparison once her cluster is operational.

Riya

  • Investigate the source of the mismatch in the direct light comparison between SKDETSIM and LUCiD, starting by increasing the number of photons in the simulation to reduce statistical variation.

Junjie

  • Report at the next meeting (in a week) on the resolution of the SIREN model's performance limitations, with the aim to finalize the paper by the end of September.

Omar

  • Provide Naomi with the branch of the LUCiD code that contains the fix for the backscatter issue once it is ready.
  • Present slides next week showing the fixes made to LUCiD and their impact on calibration.

Summary

Hyper-K Calibration Progress Report

Patrick welcomed a new participant, Naomi, who introduced herself as a PhD student working on the Hyper-Kamiokande calibration group. Naomi reported on her progress with using LUCiD to set up calibration scenarios with their light injection system, including replicating light profiles from their UK-built system. She mentioned encountering concerning amounts of backscatter at the injection site while working with LUCiD, which she is currently investigating to determine if it's a user or coding issue.

PMT Geometry Backscatter Issues

The team discussed issues with light injector positioning and backscatter in the PMT geometry. Naomi identified concerns about backscatter in her simulations, particularly with realistic light profiles compared to test sources. Omar mentioned a bug in LUCiD related to light accounting for sensors after the first step, which could explain Naomi's observations. Patrick suggested reviewing event displays for backscatter issues, and Riya reported she hadn't specifically observed backscatter problems in her work on Super-K geometry.

Beam Profile Simulation Comparison

Riya presented a comparison of beam profiles between SKDETSIM and LUCiD simulations at 405 wavelength, showing similar beam shapes but some mismatches in charge fractions. The team discussed potential causes including differences in PMT positioning and photon injection rates, with Omar noting that the current photon count of 8,000 per pulse seemed low. suggested increasing the number of photons in both simulations to reduce statistical variation and better identify the source of mismatches.

SIREN Model Parameter Discussion

The team discussed ongoing work on fixes and calibration, with one member reporting progress on slides demonstrating completed fixes that would be ready for review next week. Zhenxiong presented on probing angular response using the SIREN model with WCTE cosmic muon data at a recent collaboration meeting, though some participants questioned the approach to uncertainty propagation through SIREN parameters. The discussion revealed confusion about whether SIREN parameters have meaningful physical interpretations or if uncertainty propagation should focus on reconstructed objects rather than the model parameters themselves.

SIREN Model Physics Performance

Kazuhiro explained that when working with SIREN models, the focus should be on measuring systematics impact on final physics performance rather than individual model parameters. He suggested creating a diagram to clarify the physics goals and uncertainty measurement approach, noting that while analytical models can measure parameter uncertainties directly, SIREN models would need a different propagation method. Junjie provided an update on studying SIREN model limitations, mentioning a 2021-2022 paper that identified constraints on implicit neural representations in frequency space and the importance of the omega-0 parameter in SIREN models.

Siren Performance Issues Resolution Timeline

The team discussed a milestone paper originally due end of August, but current Siren performance issues need resolution before completion. Junjie estimated addressing the performance limits by the next meeting, with paper completion taking an additional week afterward. The group confirmed the end of September deadline remains achievable, as significant writing is already complete and only specific performance issues need resolution.