List your company
ResearchFormerly WizWand

Reinforcement Learning on Atari 20 (final)

1.2193IQM ScoreNPG-FKL
0.6620680.8067340.95141.096066Sep 11, 2025
Updated 5mo ago

Evaluation results

MethodLinks
NPG-FKLactor entropy (τ)=enabled, critic entropy (ζ)=0, actor updates (n)=12025.091.21931.16031.2827
SPMA-RKLactor entropy (τ)=enabled, critic entropy (ζ)=0, actor updates (n)=12025.091.05290.96591.1561
Adam LMCDQN (Baseline)actor updates (n)=12025.091.03630.94231.1265
NPG-RKLactor entropy (τ)=enabled, critic entropy (ζ)=0, actor updates (n)=12025.091.00920.87841.143
SPMA-FKLactor entropy (τ)=enabled, critic entropy (ζ)=0, actor updates (n)=12025.091.00880.8971.1301
DSACactor entropy (τ)=enabled, critic entropy (ζ)=0, actor updates (n)=12025.090.92260.82281.0288
DQN (Baseline)actor updates (n)=12025.090.78270.75630.8091
PPO (Baseline)actor updates (n)=12025.090.68350.62580.7311