This is the video presentation at ICML 2020 for
Being Bayesian, Even Just a Bit, Fixes Overconfidence inReLU Networks
Agustinus Kristiadi, Matthias Hein, Philipp Hennig,
International Conference on Machine Learning (ICML) 2020
Paper: https://arxiv.org/abs/2002.10118
pdf slides for this talk:
jupyter notebook with demo: https://github.com/wiseodd/last_layer_laplace