b Fdbe c J K M A = L 2 r (s, a) = 1 K K X k=1 (r✓k (s, a) ¯ r✓(s, a))2 <latexit sha1_base64="pqU3PFSYEt/dWvy83pXuiP/hL+c=">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</latexit> 1 B A1 A =: • 1 = lL hn w • 1 1 1 A hm o svT • ) 1 A pR p(✓|⌧) <latexit sha1_base64="JSwFveUnvnCQ2GwvHf4/o8PCdEY=">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</latexit> = 1: A = 2 1 n(s, a) = ¯ r✓(s, a) ¯ r ¯r <latexit sha1_base64="HHm6VJTbz+qXODjGZpRu2Wzb6jA=">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</latexit> =: dbe t J 1: = A Jy ki ac u J P Igr ✓MAP = arg max ✓ E ⌧⇠D[log p(⌧|✓) + log p(✓)] = arg max ✓ J(✓) <latexit sha1_base64="zuk3WZ8fSbJO161xMEm+hMQvhpI=">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</latexit>