Research · AWS ML Blog ·

Teaching models to forget: Selective unlearning with Amazon Nova

Teaching models to forget: Selective unlearning with Amazon Nova

AWS describes Reverse Direct Preference Optimization (rDPO), a selective unlearning technique used in Amazon Nova Customizable Content Moderation Settings. The post says rDPO reduces over-deflection while preserving model quality and includes guidance for applying similar preference optimization methods in customer exp

Read the full story at AWS ML Blog →