LEARNING-STRATEGY FOR THE BINARY PERCEPTRON

Pursuing the work of Penney and Sherrington, we determine the optimal continuous-weight perceptron which, on clipping, correctly predicts th e largest number of weights for the binary perceptron with maximum sta bility. We calculate the fraction of correctly predicted binary weight s when only the continuous weights stronger than a certain threshold a re clipped. We finally carry out simulations for a perceptron with 50 weights to test the practicability of different learning strategies.