Breaking cross-domain model protection via surrogate-guided unsupervised distribution transformation

Puntos clave

• Cross-domain models can be effectively compromised using surrogate-guided methods, showcasing significant vulnerabilities.
• The analysis found a major transformation in distribution patterns, indicating a 75% success rate in breaking protections.
• Utilizing a surrogate-guided approach, the study demonstrates a novel way of manipulating distribution in unsupervised contexts.
• These findings highlight the urgent need for enhanced protective measures in cross-domain models to prevent breaches.