We are the first to explore the connection between gradient-based adversarial attacks and the Euler method for solving ordinary differential equations (ODEs). Based on this insight, we propose a Prediction–Correction based adversarial attack framework.
For theoretical details and experimental results, please refer to our preprint: arXiv:2306.01809.

Note: This paper was originally submitted to IEEE Transactions on Information Forensics and Security (TIFS) on August 31, 2021, but was not accepted. Documentation of the submission record is provided in the uploaded file: Attachment.pdf.
The work was subsequently shared on arXiv, and as of November 11, 2025, it has received 10 citations, according to Google Scholar.

The implementation that ranks 2nd on the MadryLab MNIST White-Box Leaderboard is available in PCROS.py.