The Proximity Hypothesis
1. The Situation. An abundance of latent nections throughout the system — needed for flexibility and richness of learning possibilities.
2. Consequence. For any given function to be learned — i.e. new combination of properties — there will generally be many available latent nections.
3. Question. How is a choice made among these competitors?
4. Answer. The "winner" is the one that is, on the whole, closest to the properties being integrated. Why? Because it is the one that gets activated first, whereupon it can inhibit its competitors. (Requires definition of closest, since the properties are in different locations.) (Assumes uniform speed of transmission of activation; hence longer distance means longer time.)
5. Definition. Consider all the connections to each competitor from the properties being integrated. For each competitor, one of its connections is the longest. The closest among the competitors is the one whose longest connection from the properties being integrated is the shortest. Why? The longest connection will be the latest to activate its end point (other things being equal).
6. Consequence. The Proximity Principle: Integrating nections will tend to be maximally close to the nections for the properties they integrate.
7. Implementation. For the winning competitor to win, it must block its rivals. Such blocking can be achieved by means of inhibitory connections.
Consequences of the Learning Process and the Proximity Hypothesis
1. Learning is generally bottom-up.
2. The knowledge structure as learned by the cognitive network is hierarchical — has multiple layers.
3. An integrating nection will tend to be equidistant from the nections for the properties which it integrates. Nections for the convergence of properties of different subsystems (eg. different sensory modalities) will tend to be in locations intermediate between those subsystems.
4. Inhibitory connections should exist predominantly among nections of the same level, while excitatory connections are predominantly from one layer to the next.
5. Close competitors will tend to be neighbors — and their mutual competition is preordained even though the properties they are destined to integrate will only be established through the learning process. Therefore the presence of their mutual inhibitory connections could be genetically specified.
6. Higher levels will tend to have larger numbers of nections than lower levels.
Go to Menu (for non-frame version.)
This page was last modified on 12 September 2005.© 2000 Rice University. This document, or any portion hereof, may be used for non-commercial informational purposes only. Any copy of this document, or portion hereof, must include the copyright notice (http://www.rice.edu/about/cr-notice.html) in its entirety.