An Efficient Method of Pattern Storage in the Hopfield Net
We discuss a new a method of endowing the Hopfield net with the properties of an associative memory. A set of N patterns (biased or unbiased) may be stored in a Hopfield network of N spins with a set of connections called inverse-Hebb couplings. Furthermore, an algorithm exists called the quadratic Oja algorithm which can enhance the basin of attraction of a subset of these stored patterns. Simulations show that the combination of the quadratic Oja algorithm with initial conditions given by the inverse-Hebb rule leads to a successful alternative to the traditional Gardner algorithm. Lastly, we introduce the hardware capable of a fast implementation of the inverse-Hebb rule.
Unable to display preview. Download preview PDF.
- Coombes, S., & Taylor, J., G. (1993). Using Generalised Principal Component Analysis to Achieve Associative Memory in a Hopfield Net, Network, To Appear.Google Scholar
- Coombes, S., & Taylor, J., G. (1993). The Inverse-Hebb Rule, KCL Preprint.Google Scholar
- Jang, J., Lee, S., & Shin, S. (1988). An Optimisation Network for Matrix Inversion, Neural Information Processing Systems, Ed. D. Z. Anderson, New York, 397–401.Google Scholar