Tag
This paper presents a switching-system theory for Q-learning with linear function approximation, using joint spectral radius to analyze convergence stability under deterministic, i.i.d., and Markovian observations.