Offline RL | Deadly Triad issue

推荐阅读Why is there a Deadly Triad issue and how to handle it ?

• Bootstrapping
• Off-policy learning
• Function approximations
  当上述三者结合在一起时，value function 可能表示不稳定或者过于多样

why:

function approximation might implicitly embed Q(S, a) ≈ Q(S’, a1)