Redundancy Maximization as a Principle of Associative Memory Learning