This is a well-deserved test-of-time award. A quietly brilliant paper with an impact on how we think about deep learning that goes beyond the papers that cite it directly.