Deep Bayesian Neural Networks

• Søren Hartmann, s164182

Objectives and Project Plan

As the project is currently quite open, the specifics of the project is bound to change as the subject is explored further. An outline of objectives that I want to explore during the project can be seen below.

1. Explore classical methods for bayesian inference.
2. Find examples of bayesian deep learning in litterature.
3. Implement different inference strategies in pytorch.
4. For the different inference methods, for models of different complexity:
   1. Compare predictive performance.
   2. Evaluate how the methods handle uncertainty?
   3. How does the above compare to regular SGD models.
5. Discussion of the different methods.