A GAN-Based Defense Framework Against Model Inversion Attacks