#machine learning
Dec 2025
Non-Asymptotic Global Convergence of PPO-Clip
The theoretical foundations of the PPO-Clip algorithm are advanced by analyzing a deterministic actor-only PPO algorithm within the general RL setting with \(f\)-divergence regularization under the softmax policy parameterization and derive a non-uniform Lipschitz smoothness condition and aojasiewicz inequality for the...
Yin Liu, Qi Dai, Jun-Yu Zhang et al.
· arXiv.org · 2 citations