Adversarial risk analysis for bi-agent influence diagrams

J. González Ortega, D. Ríos Insua, J. Cano Cancela

We describe how to support the decision making of an agent (Defender) when facing an adversary (Attacker) such that their joint problem is modelled as a Bi-Agent Influence Diagram (BAID). We explore general interactions entailing sequences of both agents' decisions, some of them possibly simultaneous or repeated across time, making use of a driving example associated with Critical Infrastructure Protection (CIP). Using ID reduction operations proposed by Shachter (1986) and relevance concepts, we provide a decision adversarial risk analytic methodology to support the Defender, avoiding strong common knowledge assumptions typical of non-cooperative game theory, and we apply it to the well-known Tree Killer problem established in Koller and Milch (2003) adapted to a CIP scenario.

Palabras clave: Non-cooperative games Decision analysis Adversarial risk analysis Bi-agent influence diagram Relevance Critical Infrastructure Protection

B. Sinova Fernández, P. Terán

A. Calviño Martinez

P. Raña Míguez, G. Aneiros, J. Vilar Fernández, P. Vieu

I. García-Camacha Gutiérrez, R. Martín Martín