Hilbert matrix-based weight initialization enhanced by mutual information for neural network optimization | Synapse