(Ab)using General Search Algorithms on Dynamic Optimization Problems
2025-02-18
This blog post compares four algorithms – Bellman's principle, Dijkstra's algorithm, Monte Carlo Tree Search (MCTS), and Pontryagin's Maximum Principle – on a simple dynamic optimization toy problem. The author finds that specialized algorithms (Bellman and Pontryagin) are significantly more efficient for this specific problem, while general-purpose algorithms, while capable of finding a solution, are less efficient in terms of speed and memory usage. The post includes animations visualizing the search process of each algorithm and benchmarks comparing their performance.