a44ba9086b2b83ccf2baf7c678723449-AuthorFeedback.pdf

I.e., the 95% CIs of all mean scores are relatively small for all operators; e.g., such CIs for cartpole are< 0.3for4 each operator,much smaller than thedifferences intheir mean scores.(2)Reresultsforconstantβ,wewill Moreover, we observe that the multiplier in front ofVk(x) Qk(x,a) (i.e. We therefore introduce a family of RSOs, whereβk16 isallowed totakeonanyvalue, butitsaverage remains< 1. Furthermore, weestablish that greater variability inβk17 will lead to larger action gaps and that s.o. R1. (1) Benefit of stochasticβk is addressed by our theoretical results (Thms 3.2-3.4)