Comparison of Bayesian Neural Network Methods under Noisy Conditions

This study presents an analysis of the effects of noise on the performance of Bayesian neural networks, comparing Monte Carlo Dropout (MC-D), Bayes by Backpropagation (BBB), and Variational Inference (VI) algorithms.The research aims to identify the strengths and weaknesses of each method in order to guide practitioners in selecting the most suitable method for various applications.The performance of the methods are evaluated using uncertainty estimates, accuracy, and robustness.This study gives insights into which method works best in noisy data and how well they can estimate the noise in the data.The results show that Bayes backpropagation is the most robust method in noisy conditions, while Monte Carlo Dropout is the most accurate in noise-free conditions.Variational Inference is the most sensitive to noise, but can be used as an estimator of the noisiness of the data set