RAMP: Recognition parametrisation by Amortised Message Passing

RAMP introduces a recognition-parametrised approach for unsupervised latent-variable modelling. Instead of relying on probabilistic structures that make belief propagation tractable, or on approximations that scale poorly, it learns a flexible nonlinear amortised message-passing framework that implicitly defines latent structure. According to the abstract, RAMP supports likelihood-based recovery of latent-variable distributions in expressive nonlinear models operating on complex, high-dimensional data. The available description does not provide benchmark datasets, quantitative results, architectural details, or comparisons with existing methods.