What:
A stabilised, clipped flavour of Gradient Descent. Essentially, don’t let the new policy get too far from the old one, in a single update. PPO caps new updates if they stray too far from the old one.
A stabilised, clipped flavour of Gradient Descent. Essentially, don’t let the new policy get too far from the old one, in a single update. PPO caps new updates if they stray too far from the old one.