[month] [year]

Marimganti Srinivas

Marimganti Srinivas supervised by Dr. Ramachandra Prasad Pillutla received  his doctorate in  Computer Science and Engineering (CSE). Here’s a  summary of his research work on Quantifying Model Uncertainty using Bayesian Nonlinear Neural Networks: A Study with Sentinel-2 NDVI Data:

This research explores Bayesian methods for predicting model uncertainty in remote sensing data, focusing on two distinct approaches: Bayesian linear regression and a novel Bayesian Nonlinear Neural Network (BN3) model. Both methods utilize SENTINEL-2 Normalized Difference Vegetation Index (NDVI) data from agricultural fields in Uttar Pradesh and Madhya Pradesh, India, for time series predictions. Bayesian linear regression, a probabilistic deep
learning approach, evaluates model uncertainty through prior and posterior probability parameters using Bayesian inference. The analysis reveals high model uncertainty in predicted NDVI values, which diminishes as the data volume increases. This reduction in uncertainty is accompanied by sharper posterior density, indicating reduced variance. Regression analysis with Gaussian distribution further validates this trend, confirming that greater data availability leads to improved model certainty.
Complementing this approach, the study develops the BN3 model to address model uncertainty in neural networks. BN3 employs probabilistic modeling with variational hidden layers and output layers characterized by Gaussian distributions for variational posteriors. It assesses model uncertainty using dense variational layers with continuous, differentiable nonlinear activation functions and varying degrees of freedom. The study evaluates the goodness of fit
for nonlinear models, including polynomials and neural networks, using Root Mean Square Error as a loss function. Comparison with Deep Ensemble methods highlights the BN3 model’s effectiveness, particularly in reducing uncertainty for cubic models. Visual analyses demonstrate BN3’s consistent performance in minimizing model uncertainty across datasets, even with varying data availability. These findings demonstrate BN3’s potential as a robust tool for improving uncertainty quantification in decision support systems.

November 2024