Adversarial Machine Learning Challenges in Modern Network Security Systems

Authors:
Publisher website
Google Scholar
Abstract
In the rapidly evolving landscape of network security, the threat of adversarial attacks on machine learning models has emerged as a critical concern. In response, we propose the Adversarial Robust Network Security Framework (ARNetSec), a comprehensive defense strategy integrating three pivotal algorithms—Adversarial Training (AT), Randomized Smoothing (RS), and Defensive Distillation (DD). AT involves training the model on a mixture of clean data and adversarial examples, generated through the Fast Gradient Sign Method (FGSM). RS adds random noise to input data during inference, creating uncertainty that hampers adversarial attempts. DD employs temperature scaling in the softmax function, smoothing the probability distribution and enhancing robustness. This framework collectively fortifies machine learning models, bolstering their resistance against adversarial attacks, and substantially improving their generalization capabilities.

This publication has 19 references indexed in Scilit: