Roundtable
Alex
Looking into GX2F. Can we have a propagator that gives us also higher derivatives, so we can refine the fit without repropagating?
Doga
Looking into collider plots with Daniel also for CTD.
Luthien
Preparing slides and poster for summer school (efficient computing).
Implementing annular network this week.
Paul
Looking into shared pointer issue. Copies aren't negligible any more. Maybe add atomic ref-count for surfaces? Standard counter is 64bit, maybe we can improve there with some tricks.
Anton
Checking NNs with simpler data. Models seem to work as they should. Momentum as unit-vector made a slight improvement.
Mikkel
Started moving GNN to Athena. Athena-setup works but not reliable. Hopefully that can be finished this week.
Benjamin
Module map optimisation for CTD. Module triplets that are rotated to each other by 90 degree, it is possible to combine the statistics.
Single particle module map: Important to have high statistics with low momentum.
Background: We observed bad performance for hard scatter. ttbar-only had good performance because there were some sets in training and testing. This overestimated the quality of the fit.
Neza
TDAQ-week: plots and performance studies
Dimitra
Grouping sectors and validating Gen3. Some outliers because of simhit to LDO conversion.
Problem: Volume with 2 surfaces misses one because it goes directly to the portal.
Andreas Salzburger
Core-python bindings almost in core. code-coverage is not working yet. Continue in that direction.
Joana
Draft PR on compile time mapping for GPU-branching.
Adaptive mass tolerance for finding is stuck again. With mass tolerances surfaces start to overlap. Currently, sensitive portal-overlaps are not checked yet.
Dr. Stephen
Looking at detray unittests with sympy.
Plan to catch up on G200 developments.
Andreas Stefl
Getting breaking PRs in for next major release:
- Surface grid change to have multiple smaller cells. It was problematic for the intersections during surface finding
- Simplify intersection return type and decouple from boundaries and surface type. Should not change the performance
Plan to start to work on electron finding and also to converge to seeding2-changes.
