You are here

Modelling adaptive agents in repeated games using automata

SSP Seminar Series
Speaker: 
Francesco Figari

In the field of multi-agent learning, adaptive algorithms are becoming more and more important, while convergence to stationary equilibria is now considered a sub-optimal strategy. While stage game equilibria are still useful solution concepts, in many games they are not the best choice in terms of rewards.

The method I am developing considers a wider set of equilibria solutions, using deterministic finite automata to encode strategies, and assuming that the opponent is playing a finite set of these "regular strategies".

These automata are combined in a two-layered model, used to search for a good and stable "machine-game equilibria", while at the same time maintain the degree of freedom required to play against adaptive opponents.

Date and time: 
Tuesday, 4 November, 2008 - 12:00
Length: 
60 minutes
Location: 
IF115