Tmlr_paper

Our paper “From discrete-time policies to continuous-time diffusion samplers: Asymptotic equivalences and faster training” (with Julius Berner, Lorenz Richter, Jarrid Rector-Brooks, and Nikolay Malkin) has been accepted at Transactions on Machine Learning Research (TMLR)!
The paper proves that discrete-time GFlowNet objectives and continuous-time path-measure objectives share a single asymptotic limit, and shows how this enables faster training via coarser random discretisation.
You can find the paper here: https://openreview.net/forum?id=xLE3xJUuDO