Optimal Zero-Shot Detector for Multi-Armed Attacks

Defending signal communication from attackers is a fundamental problem in information theory (Karlof This paper explores a scenario in which a malicious and Wagner, 2003; Perrig et al., 2004). Notably, some actor employs a multi-armed attack attacks are aimed at the physical layer of the communication strategy to manipulate data samples, offering channel, which is responsible for transmitting them various avenues to introduce noise the signal. The goal of such attacks is to generate into the dataset. Our central objective is a denial of service (DoS), which involves disrupting to protect the data by detecting any alterations legitimate communication by causing intentional malfunction to the input. We approach this defensive of the communication channel (Grover et al., strategy with utmost caution, operating 2014). In a typical input perturbation scenario, a malicious in an environment where the defender possesses actor is allowed to detect and alter the signal significantly less information compared before it reaches the communication channel (Sadeghi to the attacker. Specifically, the defender is and Larsson, 2019; Tian et al., 2022). The interest in unable to utilize any data samples for training such attacks has been exacerbated by the growing popularity a defense model or verifying the integrity of machine learning (ML) models, which are of the channel. Instead, the defender relies known to be vulnerable to adversarial attacks (Goodfellow exclusively on a set of pre-existing detectors et al., 2014).