The study of learning algorithms must consider both the statistical and computational aspects of learning.
Small improvements, consistently applied, lead to massive change over time.