A picture of two adjacent layers of GLOM for a single location (i.e., part of a single column). During the forward pass, the embedding vector at level L receives input from the level L-1 embedding vector in the previous layer via a multilayer, bottom-up neural net. It also receives input from the level L+1 embedding in the previous layer via a multilayer, top-down neural net. The dependence on level L+1 in the previous layer implements top-down effects during the forward pass. The level L embedding in layer also depends on the level L embedding in layer and an attention-weighted sum of the level L embeddings at other nearby locations in layer . These within-level interactions are not shown.
This site uses cookies. By continuing to use our website, you are agreeing to our privacy policy. No content on this site may be used to train artificial intelligence systems without permission in writing from the MIT Press.