# Paul Werbos **Domain:** Machine Learning / Systems Modeling / Optimization **Doc Type:** Person Node **Maturity:** Foundational ## Definition **Paul J. Werbos** is a scientist whose 1974 Harvard dissertation developed backward derivative propagation through composed nonlinear operations and connected that machinery to adaptive systems. ## Historical Context Werbos called the procedure **dynamic feedback** and described computing derivatives by proceeding backward through an ordered table of differentiable operations. His 1981 conference contribution, published in 1982, treated backward differentiation in neural-network training. ## Sources / Provenance - Paul J. Werbos, _Beyond Regression_, 1974. - Paul J. Werbos, “Applications of Advances in Nonlinear Sensitivity Analysis,” 1982. ## See Also [[wiki/Reverse-Mode Automatic Differentiation|Reverse-Mode Automatic Differentiation]], [[wiki/Backpropagation|Backpropagation]], [[wiki/Seppo Linnainmaa|Seppo Linnainmaa]]