Accepted:
----
| |
/vol02/learning/ | © Copyright 1995 | |||
| Volume 02 | Received: Accepted: |
---- ---- |
|||
|
Learning Declarative Relations in an Oscillatory Neural Network
Stephen Bateson, James M. Hogan and Joachim Diederich |
|
| Abstract | |
| hastri and Ajjanagadde (S&A) [1] have proposed a temporal solution to the "variable binding problem" in neural networks. The work is based on the observation that synchronous, rhythmic activity occurs in the animal brain; for example, in the olfactory bulb, hippocampus and visual cortex. For instance, synchronous oscillations have been found in the visual cortex of anaesthetised cats responding to moving visual stimuli. We report results on an extension to the network proposed by S&A. The extension allows the rapid learning of instances as a result of the inference process of a temporal connectionist semantic network (CSN). The extension to the S&A system described here adds the capability of learning instances of declarative relations as a result of the temporary variable bindings that are produced by the inference process. Any such mechanism must retain the initial advantages of the S&A solution; that is, it should be parallel at the knowledge level and time for retrieval should be independent of the size of the memory system. The time for retrieval should also be independent of the number of attributes retrieved. The neural network solution suggested here has these features. | |
| Full Text |
|
|
Multimedia Links (none) Reference Links (none) Citation Reference |
Get viewers for PS & PDF   |
