Classifying by Proxy: Explainable and Reproducible Ensemble of Proxy Tasks for Child Sexual Abuse Imagery Classification
Researchers have developed a novel ensemble of proxy tasks for classifying child sexual abuse imagery (CSAI), aiming to improve reproducibility, explainability, and security. This approach, applied for the first time to real CSAI, selects relevant proxy tasks from existing literature and includes training adaptations. The final model achieved 91.9% balanced accuracy on the RCPD dataset, outperforming the representation learning model DINO and providing crucial classification explanations. AI
IMPACT Introduces a novel ensemble method for AI classification tasks, enhancing accuracy and explainability in sensitive domains.