The "binding problem" is not restricted to feedforward networks.
This problem will arise in any network model that deals with *dynamic*
relational information -- yes, even in recurrent nets, model-based nets,
and structured nets.
It is important to distinguish between "static" (or persistent) bindings
and "dynamic" (or transient) bindings. While the former can be encoded via
suitable interconnections between cells, the latter need to be encoded
in (transient) network activity states.
Synchrony is an effective and plausible means of encoding dynamic bindings
(and IMHO, *the* means used by the brain to solve this problem).
For a discussion of the binding problem, the difference between static and
dynamic bindings, and how the binding problem arises in high-level
cognition see:
Shastri, L and Ajjanagadde, V., (1993)
"From simple associations to Systematic Reasoning."
Behavioral and Brain Sciences 16(3) 417--494.
(also available at "http://www.icsi.berkeley.edu/~shastri"
look under papers on Shruti)
The above and related papers also describe a recurrent and model-based
system that shows how a temporal synchrony-based solution to the dynamic
binding problem can be harnessed to partially explain our ability to draw
rapid (reflexive) inferences (e.g., during language understanding).
-- Lokendra Shastri
Lokendra Shastri
International Computer Science Institute
1947 Center Street, Suite 600
Berkeley, CA 94704
shastri@icsi.berkeley.edu
http://www.icsi.berkeley.edu/~shastri
Phone: (510) 642-4274 ext 310
FAX: (510) 643-7684