Ensuring data security and privacy while publishing and distributing datasets.

Data Poisoning Tool

The Data Poisoning Tool embeds covert triggers within image datasets, a technique used to protect sensitive or classified visual information. Its primary application is safeguarding proprietary datasets in surveillance, reconnaissance, and security operations. For law enforcement, it helps prevent unauthorized use of image datasets, which could compromise national security, operational integrity, and privacy. The tool subtly alters images with undetectable patterns that AI systems can recognize, triggering misclassification as an indicator of unauthorized data use. This enables timely detection and response. The tool operates under a proprietary license.

Skill Level: Beginner

Watermarking data

This course will guide you through the Watermarking Tool developed by ATOS, which enhances the security and traceability of digital datasets by embedding invisible messages into images. The tool assists in tracing data ownership and downloaders, providing legal protection, and features an adjustable speed-robustness trade-off along with robustness testing against various attacks. Common attack types include compression, geometric transformations, noise addition, and filtering. The workflow involves uploading an image, inputting a message, customizing settings, watermarking the image, and testing its resilience against attacks. This tool is valuable for secure dataset sharing, ensuring data integrity, and promoting international cooperation.

Skill Level: Beginner