Evasive Active Hypothesis Testing
Meng-Che Chang, Matthieu R. Bloch · 2020
We consider an active hypothesis testing scenario in which an adversary obtains observations while legitimate parties engage in a sequential adaptive control policy to estimate an unknown parameter. The objective is for the legitimate parties to evade the adversary by controlling the risk of their test while minimizing the detection ability of the adversary, measured in terms of its error exponent. We develop bounds on the adversary's error exponent that offer insight into how legitimate adversaries can best evade the adversary's detection. We illustrate the results in a wireless transmission detection example.