TMLR paper

We got a paper published in TMLR on the connection between variational inference and generative flow networks. [Read More]

ICML paper accepted

Building on our previous work on Deep Gaussian Markov Random Fields, we – Joel Oskarsson, Per Sidén and I – got a paper accepted for the International Conference on Machine Learning (ICML). We generalize the method from lattice graphs to general graphs (turns out that it was not as straightforward as simply replacing the CNN with a GNN), propose a better variational approximation, and make the method scale to large graphs by efficient log-determinant approximations. [Read More]