Foundations Of Cognitive Science

Autoassociative Network

An autoassociative network is a connectionist system that has only a single set of processing units. These units are connected to one another (and in some architectures connected to themselves) so that changes in activity iteratively change the state of units in the network. Most autoassociative networks change network states until activation values arrive at a local energy minimum, where the entire system stabilizes. Thus unlike many connectionist networks, these networks are intrinsically dynamic, changing their states over time as a function of feedback between elements. Indeed, Ashby's (1960) Homeostat can be described as an electric autoassociative network. Some, but not all, autoassociative networks are trained with simple learning rules like the Hebb rule. Autoassociative networks can be used as memory systems (Hopfield, 1982), models of concept recognition (Anderson et al., 1977), or as models of visual processes (Dawson, 1991).


  1. Anderson, J. A., Silverstein, J. W., Ritz, S. A., & Jones, R. S. (1977). Distinctive features, categorical perception and probability learning: Some applications of a neural model. Psychological review, 84, 413-451.
  2. Ashby, W. R. (1960). Design For A Brain (Second Edition ed.). New York, NY: John Wiley & Sons
  3. Dawson, M. R. W. (1991). The how and why of what went where in apparent motion: Modeling solutions to the motion correspondence process. Psychological Review, 98, 569-603.
  4. Hopfield, J. J. (1982). Neural networks and physical systems with emergent collective computational abilities. Proceedings of the National Academy of Sciences, 79, 2554-2558.

(Added October 2009)