Track papers across labs, topics, and trends.
Search papers, labs, and topics across Lattice.
Department of Statistics & Data Science, University of Oslo. Email: geirs@math.uio.no
1
0
2
Ditch your Gaussian priors: Dirichlet scale mixture priors offer a principled way to shrink BNNs, improve robustness, and mitigate the cold posterior effect, especially when data is scarce and correlated.