Machine learningOther

Differentiable Annealed Importance Sampling Minimizes The Symmetrized Kullback-Leibler Divergence Between Initial and Target Distribution

The paper shows that Differentiable Annealed Importance Sampling (DAIS) provides more accurate uncertainty estimates than traditional variational inference by minimizing the symmetrized KL divergence.