flâneur

From Equilibrium Checking to Learning with the Open Game Engine

cybercat.institute · 1,780 words · saved by 1 readers

Compositional game theory, like game theory in general, is not just for toy models. Game theory is the standard tool for modelling in a wide range of applications in microeconomics, and many of those benefit from compositionality. One of these applications is pricing, and especially its modern version, dynamic pricing. We have been working on a research project on how to do dynamic pricing in a strategic, competitor-aware way, using compositional game theory and its natural connections to optimal control and reinforcement learning, and building on the Open Game Engine. In this post I’m going to discuss how we adapted the Open Game Engine, which is fundamentally designed as an equilibrium checker, to do multi-agent learning instead. The project we did also involved a lot of economics, which Nicolas wrote about in this post. The idea that open games can be adapted to do dynamic programming has been known since 2019, when I replicated a dynamic social dilemma environmental economics model

Compositional game theory, like game theory in general, is not just for toy models. Game theory is the standard tool for modelling in a wide range of applications in microeconomics, and many of those benefit from compositionality. One of these applications is pricing, and especially its modern version, dynamic pricing. We have been working on a research project on how to do dynamic pricing in a strategic, competitor-aware way, using compositional game theory and its natural connections to optimal control and reinforcement learning, and building on the Open Game Engine . In this post I’m going

related reading