11–16 oct. 2026
Fuseau horaire Europe/Paris

Programme Scientifique

The week combines introductory lectures, hands-on challenge work, daily mentoring, and a final plenary synthesis session. The rhythm is designed to alternate between shared learning and focused team work.

Program table

Sunday 11 October — Arrival (optional)

Time Activity
Afternoon / Evening Participant arrival and check-in
Evening Informal welcome, logistics briefing (Jean-Zay accounts, data access policies)
Evening Ice-breaker with resident artists — visual identity kick-off

Monday 12 October — Day 1: Common Ground & Problem Setting

Time Activity
Morning (plenary) Lecture 1 — Accelerators as AI systems: beam dynamics, collective effects, diagnostics, controls — where ML already works and where it fails
Morning (plenary) Lecture 2 — ML methods for scientific systems: surrogates, uncertainty quantification, physics-informed learning, anomaly detection in non-stationary systems
Morning (plenary) Lecture 3 — HPC & AI at scale (Jean-Zay): interactive workflows, data management, best practices
Afternoon Presentation of the three challenges: datasets, metrics, baselines
Afternoon Team formation — 3 groups × 2 teams (contrasting approaches)
Afternoon Hands-on: environment setup, first data exploration, baseline runs
Evening Poster-style session: "What we plan to try" — artists observe and sketch early concepts

Tuesday 13 – Thursday 15 October — Days 2–4: Hackathon Core

Each day follows a common rhythm:

Time Activity
Morning (30–45 min) Cross-group check-in: progress, blockers, short technical input
Daytime Team work: coding, model training, diagnostics, analysis
Late afternoon Mentor roundtables (accelerator physics, ML, HPC)
End of day Informal wrap-up; artist interaction with teams

Friday 16 October — Day 5: Synthesis & Confrontation

Time Activity
Morning Finalization of results; preparation of presentations and demos
Afternoon (plenary) Results session — for each group: scientific context, Team A vs Team B (assumptions, methodology, performance, interpretability), lessons learned
Closing Cross-cutting discussion: what worked / failed, transferability to real machines
Closing Outlook: publications, follow-up projects, open datasets and benchmarks
Closing Presentation of artistic output — draft comic panels
~16:00