Density Networks

David J C MacKay
Cavendish Laboratory


Density networks are latent variable models that define non-trivial densities in a space of observables. Using Bayesian inference we can discover the dimensionality of a hypothesised non-linear manifold underlying our data. I will describe a Markov chain Monte Carlo application of density networks to the modelling of protein families.
MaxEnt 94 Abstracts /