Robust Training with Adversarial Exampleson Industrial Data

In an era where deep learning models are increasingly deployed in safetycritical
domains, ensuring their reliability is paramount. The emergence of
adversarial examples, which can lead to severe model misbehavior, underscores
this need for robustness. Adversarial training, a technique aimed at
fortifying models against such threats, is of particular interest. This paper
presents an approach tailored to adversarial training on tabular data within
industrial environments.
The approach encompasses various components, including data preprocessing,
techniques for stabilizing the training process, and an exploration of diverse
adversarial training variants, such as Fast Gradient Sign Method (FGSM),
Jacobian-based Saliency Map Attack (JSMA), DeepFool, Carlini & Wagner
(C&W), and Projected Gradient Descent (PGD). Additionally, the paper delves
into an extensive review and comparison of methods for generating adversarial
examples, highlighting their impact on tabular data in adversarial settings.
Furthermore, the paper identifies open research questions and hints at future
developments, particularly in the realm of semantic adversarials. This work
contributes to the ongoing effort to enhance the robustness of deep learning
models, with a focus on their deployment in safety-critical industrial contexts.