next up previous
Next: The Graphical Model Up: Theory Previous: Prediction

Learning High-level contexts

I haven't yet described how the higher level cocepts are learned. The high level causes themselves have to be learned from the low level inputs. This could be done as follows. Lower level regions learn the most likely sequences of their inputs. After learning, whenever a part of one of those sequences occur, those regions pass up a pattern corresponding to the sequence. A higher level region with converging inputs from several lower level regions then looks at sequences occuring simultaneously in the low level regions. Patterns of sequences which consistently occur in multiple low level regions become the objects at the next higher level. This process can be repeated between levels of the hierarchy to obtain causes at the highest level. For example if region Y1 observes that the sequence sj Z1 of region Z1 and the sequence sk Z2 of region Z2 occur at the same time very often, then their combination becomes an object or cause at region Y1. Examples of learned sequences and higher level causes for line drawing movies are shown in figure 3.



Dileep George 2004-09-17