InterJournal Complex Systems, 261
Status: Submitted
Manuscript Number: [261]
Submission Date: 981228
Continuous Learning in Sparsely Connected Hopfield Nets
Author(s): Ben Goertzel

Subject(s): CX

Category: Brief Article

Abstract:

Experiments are reported in which sparsely connected Hopfield nets with capped weights perform continuous learning: they are fed an ongoing stream of training patterns, and "remember" the most recent few as attractors. Given a fixed number of neurons, the size of the sliding memory window increases as connectivity increases, so long as the weight cap is chosen appropriately.

Retrieve Manuscript
Submit referee report/comment


Public Comments: