Representing part-whole relationships in recurrent neural networks.

Our Research / Publications

janelia7_blocks-janelia7_biblio_header | block

Neural Information Processing Systems. 2005;18:563-70

Jain V, Zhigulin V, Seung HS

Note: Research in this publication was not performed at Janelia.

janelia7_blocks-janelia7_biblio_abstract | block

Abstract

There is little consensus about the computational function of top-down synaptic connections in the visual system. Here we explore the hypothesis that top-down connections, like bottom-up connections, reflect partwhole relationships. We analyze a recurrent network with bidirectional synaptic interactions between a layer of neurons representing parts and a layer of neurons representing wholes. Within each layer, there is lateral inhibition. When the network detects a whole, it can rigorously enforce part-whole relationships by ignoring parts that do not belong. The network can complete the whole by filling in missing parts. The network can refuse to recognize a whole, if the activated parts do not conform to a stored part-whole relationship. Parameter regimes in which these behaviors happen are identified using the theory of permitted and forbidden sets [3, 4]. The network behaviors are illustrated by recreating Rumelhart and McClelland’s “interactive activation” model [7].

node:body | entity_field

Main Menu (Mobile)- Block

Main Menu - Block

Representing part-whole relationships in recurrent neural networks.

Note: Research in this publication was not performed at Janelia.

Abstract