1353-UR Uncertainty Quantification in Bayesian Neural Networks for Image-Based Concentration Prediction
Location
https://campus.kennesaw.edu/colleges-departments/ccse/events/computing-showcase/fa26-cday-program.php
Document Type
Event
Start Date
20-11-2026 2:00 PM
Description
This study investigated whether a Bayesian Neural Network (BNN) could predict antigen concentrations from images and estimate predictive uncertainty under Gaussian image noise. A LeNet-5-inspired BNN was trained using 298 images from five concentration groups. Four noise settings (0, 1, 5, and 50) were evaluated using five-fold training/validation splits and a fixed test set. Prediction accuracy was measured with mean absolute error (MAE) and root mean squared error (RMSE), while uncertainty was estimated from 15 stochastic predictions per image. Results showed substantial prediction errors and variation across folds. The highest noise setting had the lowest average error and uncertainty, but the effects of noise were not consistent.
1353-UR Uncertainty Quantification in Bayesian Neural Networks for Image-Based Concentration Prediction
https://campus.kennesaw.edu/colleges-departments/ccse/events/computing-showcase/fa26-cday-program.php
This study investigated whether a Bayesian Neural Network (BNN) could predict antigen concentrations from images and estimate predictive uncertainty under Gaussian image noise. A LeNet-5-inspired BNN was trained using 298 images from five concentration groups. Four noise settings (0, 1, 5, and 50) were evaluated using five-fold training/validation splits and a fixed test set. Prediction accuracy was measured with mean absolute error (MAE) and root mean squared error (RMSE), while uncertainty was estimated from 15 stochastic predictions per image. Results showed substantial prediction errors and variation across folds. The highest noise setting had the lowest average error and uncertainty, but the effects of noise were not consistent.