CRO (clustered regression optimization) is an efficient regression algorithm for machine learning developed by William Stobaugh.
The theory behind the algorithm is that any unknown function that most reduces lost can be approximated by a large series of linear lines, at the cost of a "perfect" function, vast quantities of high dimensional data can be computed into a regression relationship without large computation needed. Additionally, the algorithm could be used to create a base equation before using gradient descent right away, potentially escaping local minima and drastically reducing computation.