Slow subspace learning from stationary processes

The talk presents a method of unsupervised learning from stationary, vector-valued processes. The method selects a subspace on the basis of an objective which can be used to bound the expected classif