The converging unlearning algorithm for the Hopfield neural network: optimal strategy
A. Yu. Plakhov · 2002
The iterative unlearning algorithm converging to the projector (pseudo inverse) rule matrix is studied in the thermodynamical limit. The system of equations describing the spectral evolution of the iterated synaptic matrix is derived. Time-dependent unlearning strength optimizing the convergence of the algorithm is found, the convergence rate being explicitly calculated.