Attacks and Defenses Against Poison and Evasion Attacks on COVID-19 Detection Models

COVID-19 detection models are crucial for accurately identifying infected individuals during the pandemic. However, these models face vulnerabilities to attacks, such as the Projected Gradient Decent (PGD) algorithm, which aims to deceive the models by generating slight modifications to input data to cause misclassification. This research investigates the PGD algorithm as an attacking technique on COVID-19 detection models and proposes effective defense strategies to counter its impact. The results highlight the importance of developing defenses against the PGD algorithm to safeguard COVID-19 detection models. By integrating Data Sanitization mechanisms, the models' ability to resist adversarial attacks is enhanced, ensuring their accuracy and reliability in identifying infected individuals. By understanding the intricacies of the PGD algorithm and implementing effective defense strategies, stakeholders can strengthen the security and trustworthiness of COVID-19 detection models. This research contributes to the development of resilient models, facilitating their reliable deployment in critical healthcare settings and public health interventions while mitigating the risks associated with adversarial attacks.