Depth-First Proof-Number Search with Heuristic Edge Cost and Application to Chemical Synthesis Planning

Akihiro Kishimoto, Beat Buesser, Bei Chen, Adi Botea

Abstract: Search techniques, such as Monte Carlo Tree Search (MCTS) and Proof-Number Search (PNS), have been proven to be effective in playing and solving games. However, understanding their performances in industrial applications is still limited. We investigate the search performance of MCTS and DFPN, a variant of PNS in Retrosynthetic Analysis. We present a new variation of DFPN that shows superior search performance over MCTS. We find that DFPN's strengths that justify its success in games have a limited value in our application domain. Therefore a MCTS variant enhanced with the formula of Segler et al. significantly outperforms DFPN. Here, we address this disadvantage of DFPN in Retrosynthetic Analysis by introducing a novel approach to combine DFPN with Heuristic Edge Initialization. We successfully demonstrate that our new DFPN search algorithm outperforms MCTS in search time by a factor of 3 on average with comparable success rates.

CMT Num: 3926