State Space Search for Risk-Averse Agents

We investigate search problems under risk in statespace graphs, with the aim of finding optimal paths for risk-averse agents. We consider problems where uncertainty is due to the existence of different scenarios of known probabilities, with different impacts on costs of solution-paths. We consider various non-linear decision criteria (EU, RDU, Yaari) to express risk averse preferences; then we provide a general optimization procedure for such criteria, based on a path-ranking algorithm applied on a scalarized valuation of the graph. We also consider partial preference models like second order stochastic dominance (SSD) and propose a multiobjective search algorithm to determine SSDoptimal paths. Finally, the numerical performance of our algorithms are presented and discussed.

• Type: Conference
• Year: 2007
• Where: IJCAI
• Authors: Patrice Perny, Olivier Spanjaard, Louis-Xavier Storme
• Keywords: Artificial Intelligence, IJCAI 2007, Multiobjective Search Algorithm, Non-linear Decision Criteria, Risk Averse Preferences