Authors:
(1) Ephrem Bloom, Toyota Technology Institute in Chicago, Illinois, USA;
(2) Melissa Duts, Toyota Technology Institute in Chicago, Illinois, USA.
Links table
Summary and one introduction
2 preparation and 2.1 models of behavioral opponents
3 introductions and intuition
4.1 Best short -sighted response and 4.2 gambling
4.3 Winning, staying and losing the shift to the opponent
4.4 The opponent that follows the leader and 4.5 The opponent has the highest average reward
5 circular
5.1 Other behavioral strategies
5.2 Exploiting an unknown strategy from a well -known group of strategies
6 future work and references
extension
A.1 Variable victory, survival, loss and transformation: tie and survival
A.2 The Leader’s follow -up variable: limited date
A.3 The limits of elliptical errors
A.4 is the highest average return for discount
4.3 Winning, staying and losing the shift to the opponent
Remember that the Win-Stay Lose-Shift plays the same procedure immediately after winning, and turns into the next procedure in his procedure arrangement immediately after the loss. The TIE-Shift variable for this opponent deals with a tie as a loss and transformations, while the TIE-STAY variable deals with a tie as a victory and remains.
4.3.1 alternative: tie transformation
guide. In the first stage, we record the correct procedure arrangement: The opponent begins to play the first procedure in the arrangement of its procedure and always moves to the next procedure in the arrangement, so by noticing the N -1 transformations we notice all the procedures N in the correct order request. Since we play every consecutive action against the current procedure of the opponent, we guarantee them the shift after N -1 tours at most, since their procedure should be linked to itself and lose in front of at least one other procedure.
At the beginning of stage 4, we expect the following procedure for the opponent correctly: We know that we have won the last round, so we know that the opponent will move to the next procedure in the arrangement of his procedure (which we recorded correctly, as shown above). The above have shown that we have properly registered the best response to each action, so we win by running the best registered response to the expected procedure. At the beginning of the next round, the same conditions remain (and will continue after each round), so we will win all subsequent rounds.
4.3.2 alternative: tie tie
The main difference is to overcome the TIE-Stay variable compared to the Tie-Shift variable that we can find the best response to each action once each action is played in a row until the discount turns, as it only turns after the loss. For algorithm, theory and proof, see A.1 in the supplement.
