J. González Ortega, D. Ríos Insua, F. Ruggeri, R. Soyer
There has been an increasing interest in Hypothesis Testing (HT) problems in which hostile adversaries perturb the data observed by a decision maker to confound her about the relevant hypothesis and gain some benefit, e.g. in adversarial classification and adversarial machine learning. Most solutions have focused on game theoretic approaches to HT, with the entailed common knowledge assumptions. However, this is not realistic since the adversaries’ beliefs and preferences will typically not be readily available. Using Adversarial Risk Analysis (ARA), we provide an alternative novel approach to the Adversarial Hypothesis Testing (AHT) problem. We study the bi-agent AHT case from the defender's perspective, formulating a Bayesian decision making problem which requires to forecast the attacker's decisions by simulating his own problem, acknowledging our uncertainty over his beliefs and preferences. An application in relation with batch acceptance is used for illustrative purposes.
Palabras clave: Statistical decision theory, Bayesian analysis, Adversarial risk analysis, Security, Batch acceptance
Programado
GT8-1 Inferencia Bayesiana
4 de septiembre de 2019 10:40
I2L7. Edificio Georgina Blanes