Close

Presentation

SpecMCTS: Accelerating Monte Carlo Tree Search using Speculative Tree Traversal
TimeTuesday, December 7th6:00pm - 7:00pm PST
Event Type
Networking Reception
Work-in-Progress Poster
Virtual Programs
Presented In-Person
DescriptionMonte Carlo Tree Search (MCTS) algorithms show great success in decision-making problems. MCTS requires significant computing loads to make a good decision. Parallelizing MCTS is challenging because MCTS is a sequential process. We present SpecMCTS, which accelerates MCTS by speculatively traversing the search tree. SpecMCTS uses a pair of DNN models, the speculation model and the main model. The faster (but less accurate) speculation model accelerates the tree search while the more accurate main model improves the decision quality. SpecMCTS accelerates MCTS for the game of Go by up to 3.81x while having better decision quality than previous approaches.