Strategic decision making under large amounts of hidden information

Develop a practical foundation for strategic decision making in imperfect-information settings with large amounts of hidden information, such as the board game Stratego, that overcomes the scaling limitations of public-information-based problem transformations and enables effective decision making when explicit enumeration is infeasible.

Background

The paper discusses the inherent difficulty of imperfect-information interactions, where agents lack access to strategically relevant information. In such settings, decision values depend on complex interdependencies among contemporaneous and historical counterfactuals, making it nontrivial to determine optimal action frequencies and strategies.

Prior successful methods rely on transformations that operate on public information, but the computational cost of these techniques scales with the amount of hidden information. This has limited their applicability to domains with relatively small hidden-information spaces (e.g., Texas hold'em), leaving larger domains like Stratego—characterized by an astronomically large space of possible hidden piece configurations—as outside practical reach.

The authors explicitly identify the lack of an alternative practical foundation, combined with the scaling limitations of public-information-based methods, as a reason why strategic decision making in settings with large hidden information has remained an open problem. This statement sets the stage for their contributions aimed at addressing these limitations in Stratego.