Home » Research » Research groups » Neuroinformatics » Statistical theory » Score matching

Score matching

One often needs to estimate statistical models where the probability density function is known only up to a multiplicative normalization constant. While we encounter this problem in our statistical models of visual processing, it is, in fact, a very general problem in statistical estimation.

Typically, one then has to resort to Markov Chain Monte Carlo methods, or different kinds of approximations. We have proposed a new method that is computationally very simple yet statistically consistent, based on matching the score functions of the model and data densities.

For publications on score matching, see this page.