AnalysisAI ModelsSeptember 17, 2026

FRAUDSkill freezes audio-language model weights for anti-fraud detection

Read original source →arxiv.org

FRAUDSkill targets audio anti-fraud detection with large audio-language models by keeping weights frozen and optimizing structured skills instead of fine-tuning. The method constrains predictions to a predefined label space and a structured decision protocol built around service scenarios.

1 source

More stories today

Open the live feed