Randomization Defense

Dropout Defense

Strategic use of dropout during inference (normally disabled) to introduce randomness into network activations, making gradient-based attacks less reliable.

← 뒤로