Speaker
Description
The study investigates the use of Bayesian Neural Networks (BNNs) to estimate aleatoric and epistemic uncertainties in radiation‑dose measurements obtained with Poly(Allyl Diglycol Carbonate) solid‑state nuclear track detectors. The detectors were irradiated with alpha particles and fast neutrons across multiple experimental configurations, producing a complex dataset well-suited for machine‑learning analysis. A representative experimental configuration was selected as a baseline for model development. Within a unified probabilistic framework, the BNN approach quantifies uncertainty arising from both model parameters and intrinsic measurement noise, and these contributions are visualized to illustrate their behavior. Detector responses were processed using the commercial TASLImage system, and the resulting TASLImage dose predictions with associated BNN‑based uncertainties were compared against the true experimental dose values to assess model performance.