Analýza, Detekce a Prevence Kybernetických Hrozeb v Prostředí Strojového Učení

The bachelor's thesis stems from the growing need to address security threats in the environment of Machine Learning (ML) and Deep Learning (DL) models, focusing on the identification and prevention of cyber threats that could impact the reliability and credibility of these models. The aim of the thesis is to systematically analyze security risks, weaknesses, and threats, and propose methods for their detection and prevention, using an experimental ML model predicting diabetes, which is tested for the effectiveness of three selected cyber attacks. The results of the experimental research demonstrate the real impacts of these attacks on the model. Subsequently, potential weak points are identified based on the results, and protection, detection, and prevention methods against attacks on ML models are proposed. Ethical and legal aspects of these attacks are also considered. This thesis highlights the importance of researching the security of ML and DL models, providing a comprehensive overview of potential threats and their impacts. Thus, it contributes to a better understanding of security risks in the field of ML and DL and helps develop more effective strategies for protecting these models in practice.