Jul 2026
Parallel Noising in Neural Markov Logic Networks
A new training and inference algorithm inspired by parallel-tempering Markov chain Monte Carlo methods is developed, which is named parallel noising and allows NMLNs to match the performance of specialized text-based recurrent models when generating small molecular structures.
Peter Jung, Giuseppe Marra, Ondřej Kuželka
· arXiv.org · 0 citations