October 11th at 17:30, in ILLC Seminar Room (F1.15)
When is information useful? How useful is it?
In this talk, I will propose a reasonable way of measuring the relevance of a piece of information to an agent. Specifically, I will suggest that relevance should be measured in terms of the expected change in the logarithm of the agent's capital.
This definition has an interpretation in terms of economic behavior, and it coincides with classical Shannon information under certain nice conditions. Under not-so-nice conditions, it sheds some new light on the value of information in various oddball betting situations.
The paper that this talk will be based on can be read here.