Introduction to Bayesian Neural Networks.
Speaker: Tatsuoka Caroline (OSU)
Abstract: I will be reviewing the Bayesian Neural Network framework. With sufficient data, current deep learning technologies offer a tool to better understand and predict dynamical systems behavior. However, uncertainty in predictions can arise due to a lack of data and from the stochastic nature of the optimization algorithms. This can result in a lack of robustness and interpretability among prediction values. BNN’s offer a probabilistic approach for such uncertainty quantification. I will review some papers which touch on some BNN structures, which include sampling and variational inference methods. A numerical example applied to a simple differential equation is generated to demonstrate the potential advantages and challenges of this framework.
Comments are closed, but trackbacks and pingbacks are open.