FSE+Attention model for particle identification in ALICE (Pb-Pb Run 3) at CERN. State-of-the-art detector masking with 92.8% accuracy using JAX/Flax. Handles missing detector data through Feature Set Embedding + Multi-head Attention. Production-ready with Focal Loss, class weighting, and two-tier model persistence.
machine-learning deep-learning neural-network flax missing-data high-energy-physics particle-physics cern class-imbalance attention-mechanism alice-experiment particle-identification focal-loss jax feature-embedding heavy-ion-physics detector-physics detector-masking
-
Updated
Nov 24, 2025 - Jupyter Notebook