A Latent-Variable Grid Model
Rajasekaran Masatran · arXiv (Cornell University) · 2015
A major problem with the markov random field (MRF) is that its learning algorithms are computationally expensive. Grid associative MRFs occur frequently in sequence and image learning. We design a non-markov high-bias low-variance model as an alternative to this subclass of MRF. Our learning algorithm uses vector quantization and, at time complexity O(T^d log T^d), is significantly faster than that of MRF.