Empirical analysis of recent stochastic gradient methods for approximate inference in Bayesian deep learning, including SWA-Gaussian, MultiSWAG, and deep ensembles. See report_localglobal.pdf.
This is my master's research project, supervised by Hippolyt Ritter and David Barber at the UCL Centre for AI. If interested I recommend reading the introductory chapter of my report (report_localglobal.pdf) to get a high-level idea.
https://github.com/wjmaddox/swa_gaussian
Empirical analysis of recent stochastic gradient methods for approximate inference in Bayesian deep learning, including SWA-Gaussian, MultiSWAG, and deep ensembles. See report_localglobal.pdf.