Regret in the on-line decision problem (Q1818283)

The paper deals with a specific type of the decision making sequences. It assumed that a decision-maker makes a time sequence of decisions, and his final loss depends on the accepted decision as well as on the actual state of the world. Moreover, in the time of the decision making the agent has no information about the state of the world. His aim is to minimize the total loss. A basic concept of the presented model is the one of the ``regret'' which characterizes the difference between the loss following after some decision, and the expected loss connected with an alternative decision making strategy. The main results formulate the properties of the regret and its relation to the optimal decision making.