Damage Resistance

Damage resistance is one of the putative advantages of parallel distributed processing models (Dawson, 2004).  In such models, knowledge is “smeared” throughout many connections, indicating that there is tremendous redundancy in knowledge representation.  One consequence of this is that damage to a small number of connections in such a system will decrease, but not destroy, the system’s performance.  This is opposite to the notion of brittleness (a characteristic of classical models), in which damage to a component has catastrophic effects.  Damage resistance requires nonlocal distributions.  When a component is ablated from a connectionist network, the more local its functional role, the greater is the (negative) impact on system performance (Medler, Dawson & Kingstone, 2005).

References:

1. Dawson, M. R. W. (2004). Minds And Machines: Connectionism And Psychological Modeling. Malden, MA: Blackwell Pub.
2. Medler, D. A., Dawson, M. R. W., & Kingstone, A. (2005).  Functional localization and double dissociations: The relationship between internal structure and behavior. Brain and Cognition, 57, 146-150.

(Added April 2011)