Adversarial Training Methods for Deep Learning: A Systematic Review

Deep neural networks are exposed to the risk of adversarial attacks via fast gradient sign method (FGSM), projected descent (PGD) attacks, and other attack algorithms. Adversarial training is one methods used defend against threat attacks. It a schema that utilizes an alternative objective function provide model generalization for both data clean data. In this systematic review, we focus particularly on as improving defensive capacities robustness machine learning models. Specifically, sample accessibility through generation methods. The purpose review survey state-of-the-art robust optimization identify research gaps within field applications. literature search was conducted using Engineering Village (Engineering engineering tool, which provides access 14 patent databases), where collected 238 related papers. papers were filtered according defined inclusion exclusion criteria, information extracted from these strategy. A total 78 published between 2016 2021 selected. Data categorized strategy, bar plots comparison tables show distribution. findings indicate there limitations optimization. most common problems overfitting.