# Adversarial Robustness Training

A CLIP model fine-tuned with sparse autoencoder (SAE), excelling in zero-shot image classification tasks, particularly adept at recognizing adversarial typographic attacks