Dynamic Programming

Dynamic programming uses Bellman recurrences to break a large sequential decision or optimization problem into smaller subproblems whose solutions can be reused. Exact methods—including memoization, state-space pruning, constraint propagation, and memory-saving algorithms such as Hirschberg’s—are used in scheduling, clustering, speech alignment, and control; approximate methods use neural value functions, sampling, or Monte Carlo tree search for larger problems. Researchers evaluate these approaches on settings such as stochastic control, dynamic pricing, combinatorial benchmarks, and long speech sequences. A central tension is preserving solution quality and guarantees while controlling the time, memory, and state-space costs that can make exact methods impractical.