推荐andreyk · 2021-12
“More recently there was EfficientZero, an extension to MuZero, which is pretty cool as it is both sample efficient and quite performant”
推荐 EfficientZero,作为 MuZero 的扩展,样本效率高且表现好
EfficientZero:基于 MuZero,仅两小时游戏经验即达人类水平
提出基于 MuZero 的样本高效的基于模型视觉强化学习算法 EfficientZero。在 Atari 100k 基准上,仅用两小时真实游戏经验就达到 194.3% 的平均人类水平和 109.0% 的中位数,性能接近 DQN 在 2 亿帧下的表现,所用数据少 500 倍;论文称这是首次在如此少数据下于 Atari 上达到超人表现。
评论者把它作为 MuZero 的后续扩展推荐,称其样本效率高、性能好,首次在 Atari 100k 上达到超人水平且数据比 DQN 少 500 倍。
“More recently there was EfficientZero, an extension to MuZero, which is pretty cool as it is both sample efficient and quite performant”
“Mastering Atari Games with Limited Data: Mastering Atari Games with Limited Data”
“Have you heard about EfficientZero? This is the first algorithm that achieved super-human performance on Atari 100k actions benchmark.”