×

Sleep refreshed memory for neural network

  • US 4,926,064 A
  • Filed: 07/26/1989
  • Issued: 05/15/1990
  • Est. Priority Date: 07/22/1988
  • Status: Expired due to Fees
First Claim
Patent Images

1. A neural network for use with a learning algorithm, a sleep refresh algorithm, and a plurality of environmental input signals, each of said environmental input signals having an activity level, comprising:

  • a plurality of N neuronal outputs, each carrying a signal having an activity level, said plurality including M first level neuronal outputs and N-M second level neuronal outputs;

    feed-forward response means for generating on a j'"'"'th one of said second level neuronal outputs, M<

    j≦

    N, a j'"'"'th signal having an activity level responsive to the temporal and spatial sum of the activity levels of said signals on all i'"'"'th ones of said first level neuronal outputs, i=1,. . . ,M, each as weighted by a respective connectivity weight wij ;

    means for modifying at least one of said connectivity weights wij in accordance with said learning algorithm; and

    sleep activated means for, during a sleep period, isolating said neuronal outputs from said environmental inputs and adjusting at least some of said connectivity weights wij according to said sleep refreshed algorithm.

View all claims
  • 1 Assignment
Timeline View
Assignment View
    ×
    ×