Yes, label propagation is very similar to belief propagation on a Gaussian Markov random field (a particularly simple type of graphical model). To be precise th

Yes, label propagation is very similar to belief propagation on a Gaussian Markov random field (a particularly simple type of graphical model). To be precise this is 'inference' on the graphical model, rather than 'training' or 'learning'. See the author's 2016 JMLR paper for more details: http://www.jmlr.org/proceedings/papers/v51/ravi16.pdf http://www.jmlr.org/proceedings/papers/v51/ravi16.pdf