Bayesian Validation Shows Diffusion Models Reconstruct Missing Calorimeter Data in Heavy-Ion Collisions
Pretrained diffusion models serve as Bayesian priors for inpainting incomplete calorimeter data in heavy-ion collisions. Systematic posterior diagnostics confirm calibrated energy response and spatial reconstruction across varying mask sizes and centralities. The validation framework provides a general template for probabilistic treatment of detector inefficiencies.
A key limitation is reliance on simulation-trained priors whose fidelity to real detector response remains untested on experimental data. The paper reports results only on Monte Carlo samples; closure tests on actual collision events with artificially masked regions will be required to confirm transferability. Follow-up work embedding the inpainting step inside full unfolding frameworks could quantify downstream impact on extracted jet or flow observables.
Roli Esha: Within 12 months, application of the diffusion inpainting pipeline to real LHC Run 3 Pb-Pb data will reduce jet-energy-scale systematic uncertainty by at least 8% in events with >15% masked calorimeter area.
Sources (2)
- [1]Primary Source(https://arxiv.org/abs/2608.14801)
- [2]Supporting Source(https://arxiv.org/abs/2305.13389)