Computational Game Theory 8.1 External Regret -reminder

The model (single player): • Actions A = {a1 . . . am} • Time dependent loss function, (can also be defined as a gain function): l i loss of action ai at time t The game: a. For each step t, the player chooses a distribution p ∈ ∆(A) b. The adversary decides on the losses l = (l 1 . . . l t m), where l t i ∈ [0, 1] c. The player’s loss at time t is l ON = ∑m i=1 p t il t i . The cummulative loss up to time T is