Speeding up Stochastic Dynamic Programming with Zero-Delay Convolution

We show how a technique from signal processing known as zero-delay convolution can be used to develop more efficient dynamic programming algorithms for a broad class of stochastic optimization problems. This class includes several variants of discrete stochastic shortest path, scheduling, and knapsack problems, all of which involve making a series of decisions over time that have stochastic consequences. We also correct a flaw in the original analysis [8] of the zero-delay convolution algorithm.