Adversarial attack defense analysis: An empirical approach in cybersecurity perspective
Kousik Barik, Sanjay Misra
Abstract
Advancements in artificial intelligence in the cybersecurity domain introduce significant security challenges. A critical concern is the exposure of deep learning techniques to adversarial attacks. Adversary users intentionally attempt to mislead the techniques by infiltrating adversarial samples to mislead the prediction of security devices. The study presents extensive experimentation of defense methods using Python-based open-source code with two benchmark datasets, and the outcomes are demonstrated using evaluation metrics. This code library can be easily utilized and reproduced for cybersecurity research on countering adversarial attacks. Exploring strategies for protecting against adversarial attacks is significant in enhancing the resilience of deep learning techniques. • Empirical Analysis: Conducted extensive experiments on defense methods against adversarial attacks. • Open-Source Tools: Developed and shared Python-based code for replicable research. • Benchmark Datasets: Utilized two benchmark datasets for robust testing. • Evaluation Metrics: Demonstrated outcomes using clear evaluation metrics. • Enhanced Resilience: Advanced strategies to improve deep learning defenses in cybersecurity.