A New Approach to the Snake-In-The-Box Problem

摘要

The "Snake-In-The-Box" problem, first described more than 50 years ago, is a hard combinatorial search problem whose solutions have many practical applications. Until recently, techniques based on Evolutionary Computation have been considered the state-of-the-art for solving this deterministic maximization problem, and held most significant records. This paper reviews the problem and prior solution techniques, then presents a new technique, based on Monte-Carlo Tree Search, which finds significantly better solutions than prior techniques, is considerably faster, and requires no tuning.