Approximate Inference; Convolutional Neural Networks; Stein's Method

Approximate inference algorithms [Slides]

Souvik Chakraborty

Convolutional Neural Networks [Slides]

Navid Shervani-Tabar

Intro to Stein's method, with applications to Bayesian Surrogate Modeling [Slides]

Yinhao Zhu

Stein's method [1] is a theoretical technique to obtain bounds on the distances between probability distributions. It was recently combined with kernel methods and applied to Bayesian inference [2]. This talk reviews the basics of this recent development and shows an application to Bayesian neural net as a Bayesian surrogate for UQ [3].

References:

[1] https://sites.google.com/site/steinsmethod/

[2] Liu, Qiang, and Dilin Wang. "Stein variational gradient descent: A general purpose bayesian inference algorithm." Advances In Neural Information Processing Systems. 2016.

[3] Bilionis, Ilias, and Nicholas Zabaras. "Multi-output local Gaussian process regression: Applications to uncertainty quantification." Journal of Computational Physics 231.17 (2012): 5718-5746.