Research3h ago
Apple Proposes Method to Lock Model Weights
A new technique from Apple uses deep low-rank residual distillation to prevent unauthorized fine-tuning of open-weight models. The method locks pretrained weights to stop malicious adaptations. While most developers prefer open checkpoints, this research targets high-security deployments. It offers a technical deterrent against weight modification for practitioners managing sensitive model releases.