main.out

\BOOKMARK [0][-]{section*.2}{Model-free and Bayesian Ensembling Model-based Deep Reinforcement Learning for Particle Accelerator Control Demonstrated on the FERMI FEL}{}% 2
\BOOKMARK [1][-]{section*.1}{Abstract}{section*.2}% 1
\BOOKMARK [1][-]{section*.3}{Introduction and Motivation}{section*.2}% 3
\BOOKMARK [2][-]{section*.5}{An Overview of the Main Results}{section*.3}% 4
\BOOKMARK [1][-]{section*.6}{The set-up of the studied problem}{section*.2}% 5
\BOOKMARK [2][-]{section*.7}{The Physical Set-up}{section*.6}% 6
\BOOKMARK [2][-]{section*.8}{The Training Environment}{section*.6}% 7
\BOOKMARK [1][-]{section*.10}{Deep Reinforcement learning}{section*.2}% 8
\BOOKMARK [2][-]{section*.11}{Model-free Reinforcement Learning}{section*.10}% 9
\BOOKMARK [3][-]{section*.12}{Approximate dynamic programming}{section*.11}% 10
\BOOKMARK [3][-]{section*.13}{Design decisions for MFRL}{section*.11}% 11
\BOOKMARK [2][-]{section*.14}{Uncertainty Aware DYNA-style Reinforcement Learning}{section*.10}% 12
\BOOKMARK [2][-]{section*.16}{Critical Design Decisions in MBRL}{section*.10}% 13
\BOOKMARK [3][-]{section*.17}{The uncertainty aware dynamics model}{section*.16}% 14
\BOOKMARK [3][-]{section*.18}{The controller algorithm}{section*.16}% 15
\BOOKMARK [3][-]{section*.19}{Handling of the uncertainty}{section*.16}% 16
\BOOKMARK [3][-]{section*.20}{The data acquisition}{section*.16}% 17
\BOOKMARK [1][-]{section*.21}{Experimental results from FERMI RL online tests}{section*.2}% 18
\BOOKMARK [2][-]{section*.23}{MFRL Tests}{section*.21}% 19
\BOOKMARK [2][-]{section*.29}{MBRL Tests}{section*.21}% 20
\BOOKMARK [1][-]{section*.34}{Discussion and outlook}{section*.2}% 21
\BOOKMARK [1][-]{section*.36}{Conclusions}{section*.2}% 22
\BOOKMARK [1][-]{section*.37}{Acknowledgements}{section*.2}% 23
\BOOKMARK [1][-]{section*.38}{A Non-linear Standard Control Problem}{section*.2}% 24
\BOOKMARK [2][-]{section*.39}{NAF2 Details}{section*.38}% 25
\BOOKMARK [2][-]{section*.41}{The Impact of Noise}{section*.38}% 26
\BOOKMARK [3][-]{section*.43}{Regression assuming homoskedastic Gaussian noise using `anchored ensembling'}{section*.41}% 27
\BOOKMARK [2][-]{section*.46}{NAF versus AE-DYNA}{section*.38}% 28
\BOOKMARK [1][-]{section*.48}{References}{section*.2}% 29