Evaluating Anytime Algorithms for Learning Optimal Bayesian Networks
This work addresses computational bottlenecks in Bayesian network learning for AI and data science applications, though it appears incremental as it adapts existing algorithms rather than introducing fundamentally new paradigms.
The researchers tackled the problem of learning optimal Bayesian networks when exact algorithms fail due to resource constraints by adapting anytime heuristic search-based algorithms, finding that the anytime window A* algorithm typically discovers higher-quality or optimal networks faster than other methods.
Exact algorithms for learning Bayesian networks guarantee to find provably optimal networks. However, they may fail in difficult learning tasks due to limited time or memory. In this research we adapt several anytime heuristic search-based algorithms to learn Bayesian networks. These algorithms find high-quality solutions quickly, and continually improve the incumbent solution or prove its optimality before resources are exhausted. Empirical results show that the anytime window A* algorithm usually finds higher-quality, often optimal, networks more quickly than other approaches. The results also show that, surprisingly, while generating networks with few parents per variable are structurally simpler, they are harder to learn than complex generating networks with more parents per variable.