핵심 개념
Local notions of correlated equilibria in non-concave games are linked to projected gradient dynamics, leading to approximable equilibria.
초록
The content explores local correlated equilibria in non-concave games, focusing on approximations and performance guarantees. It delves into the concept of coarse equilibria and their relation to gradient dynamics. The analysis covers the tractable computability of equilibria, extending to stationary and local correlated equilibria. Key insights include the role of smooth boundaries and Lipschitz continuity in approximating equilibria.
통계
As a result, such equilibria are approximable when all players employ online (projected) gradient ascent with equal step-sizes as learning algorithms.
For (1), the approximation bound decreases in K, while the class of polyhedra considered in (2) contain the simplex and the hypercube as special cases.
인용구
"Our analysis shows that such equilibria are intrinsically linked to the projected gradient dynamics of the game."
"We identify the equivalent of coarse equilibria in this setting when no regret is incurred against any gradient field of a differentiable function."