Train the Model, Not the Reader: Decodability Supervision for Verifiable Activation Explanations

Part of

Multimedia Misinformation Detection Advances