AID2E bi-weekly technical meeting
Meeting: February 19, 2026
Attendees: Wen Guan, Tianle Wang, Kolja Kauder, Baptiste Fraisse, Tanja Horn, Cynthia Nunez, Derek Anderson, Cristiano Fanelli, Karthik Suresh, Hemalata Nayak
- Announcements: none.
- Phase I
- Round robin:
- Derek: discussing BIC design dimensions and objectives, including angular resolution for photons.
- Wen: pushed commits to AID2E-scheduler. Joblib scheduler works well. PanDA runner is under test.
- DAG executor now supports schedulers.
StageStatus now tracks the PanDA task status.- Execution hierarchy clarified as branches, workflows, stages, and jobs.
- Separate pull request planned for these changes.
- BIC can be the focus for workflow examples.
- Karthik:
- Ax-based optimization is integrated and under validation.
- PyMOO work is focused on reconciling NumPy arrays with tensor-based handling in AID2E.
- dRICH (Duke): optimizing the triple point. A quartz window is implemented and usable for optimization. More focus is currently on IRT2.
- Paper timeline:
- Target for AID2E-scheduler completion was Feb 28, 2026.
- Overleaf draft: project link
- Checklist includes schedulers, DAG executor, tests, helper APIs, ePIC workflow examples, algorithms, and problems.
- PanDA token lifetime is still short and needs periodic renewal.
- DTLZ2 plots may need to be regenerated in AID2E-framework rather than relying only on Dec 2025 results.
- Cris suggested placeholders first, then replacement with updated plots.
- Minimal paper set: one test problem and one real workflow, with BO and optionally EA.
- Karthik suggested constrained DTLZ problems for broader algorithm testing.
- Placeholder plots targeted for Feb 24, 2026.
- ePIC-related plots should avoid physics performance claims.
- Narrative for the paper should be drafted soon.
- Potential journals include Computing Software and Big Science and Computer Physics Communications.
- Phase II
- Cynthia shared a survey over email: survey
- Wen presented a PanDA RAG and MCP direction using LangGraph.
- The group discussed scope, framework choice, and comparing multiple approaches before deciding.
- Closure-test problem definition using natural language was deferred to the next meeting.
- AOB: none.