Analyzing Local Equilibria in Non-Concave Games
The authors propose a new solution concept, termed (ε, Φ(δ))-local equilibrium, to address challenges in non-concave games and show convergence guarantees of Online Gradient Descent and no-regret learning.