Assessing the Vulnerability of Machine Learning Models to Cyber Attacks and Developing Mitigation Strategies
Abdul Sajid Mohammed, Shubham Jha, Ayisha Tabbassum, Vaibhav Malik · 2024
The integration of machine learning (ML) models into various sectors has revolutionized industries by enabling advanced data analytics, pattern recognition, and decision-making processes. However, the increasing adoption of ML technologies also raises concerns about their vulnerability to cyber-attacks. Adversarial attacks, data poisoning, and model inversion are among the various tactics employed by threat actors to compromise the integrity, confidentiality, and availability of ML systems. This paper critically assesses the vulnerabilities of ML models to cyber threats and explores effective mitigation strategies. Through a comprehensive literature review, common attack vectors, vulnerabilities, and mitigation techniques are identified and analyzed. Adversarial training, robust optimization, and input sanitization are among the strategies examined for enhancing the security and resilience of ML models. The study underscores the importance of collaboration and knowledge sharing in developing effective defense mechanisms against emerging threats in ML security.