2718. Adversarial Risk Analysis with Fully Probabilistic Designs for sequential games
Invited abstract in session MD-33: Algorithms in Decision Modelling, stream Decision Analysis.
Monday, 14:30-16:00Room: Maurice Keyworth 1.31
Authors (first author is the speaker)
| 1. | Mario Chacón-Falcón
|
| Datalab, Institute of Mathematical Sciences | |
| 2. | Tatiana Guy
|
| Adaptive Systems, UTIA, Czech Academy of Sciences | |
| 3. | Miroslav Kárný
|
| Adaptive Systems, UTIA, Czech Academy of Sciences | |
| 4. | David Rios Insua
|
| Royal Academy of Sciences of Spain |
Abstract
Adversarial risk analysis (ARA) is a decision analytic methodology that informs decision-making when facing intelligent opponents and uncertain outcomes. It enables an analyst to model beliefs about an opponent's utilities, capabilities, probabilities, and the strategic calculations that an opponent employs. A major aplication of ARA is in defend-attack games, mitigating standard common knowledge and common prior assumptions in standard game-theoretic settings. Yet ARA entails very involved adversarial modeling and computations. To mitigate these, we explore here the combination of ARA with fully probabilistic designs (FPD), used in control theory to support an agent's decisions in choosing the decision distribution that is closest to the agent’s ideal trajectory. Specifically, we use FPD to model and forecast the opponent's decisions within an ARA setting and present how to handle simple defend-attack template games. We also discuss how to merge ARA and FPD in sequential games to handle reinforcement learning problems under threats.
Keywords
- Decision Analysis
- Risk Analysis and Management
- Game Theory
Status: accepted
Back to the list of papers