Modeling image patches with a directed hierarchy of Markov random fields Simon Osindero & Geoffrey Hinton
· A new learning algorithm for a directed hierarchy of MRF's.
Undirected Connections

Poster ID M56

· MRF's in the hidden layers allow it to model image patches very well. · The learning algorithm is efficient and scalable because it learns one hidden layer at a time. · Inference in the learned model is very fast and simple.

Hidden MRF
Directed Connections

Hidden MRF
Directed Connections

Data