notesum.ai

Published at December 3

Conformal Symplectic Optimization for Stable Reinforcement Learning

cs.LG

cs.AI

Released Date: December 3, 2024

Authors: Yao Lyu¹, Xiangteng Zhang¹, Shengbo Eben Li¹, Jingliang Duan², Letian Tao¹, Qing Xu¹, Lei He¹, Keqiang Li¹

Aff.: ¹Tsinghua University; ²University of Science and Technology Beijing

Arxiv: http://arxiv.org/pdf/2412.02291v1

Refer to caption

Optimizer	RAD	ADAM	$\Uparrow$
Breakout-v4	228 $\pm$ 41	125 $\pm$ 40	82.4%
Enduro-v4	1509 $\pm$ 245	929 $\pm$ 63	62.4%
Seaquest-v4	5871 $\pm$ 835	2301 $\pm$ 247	155.1%
SpaceInvaders-v4	730 $\pm$ 122	479 $\pm$ 83	52.4%