The paper introduces a novel defense mechanism against adversarial attacks in image classification, called High-Level Representation Guided Denoiser (HGD). Traditional denoising methods suffer from the error amplification effect, where small adversarial perturbations are amplified to large errors in the model's output. HGD addresses this issue by using a loss function that measures the difference between the target model's outputs for clean and denoised images. This approach is more robust to both white-box and black-box adversarial attacks compared to ensemble adversarial training, which is the current state-of-the-art method. HGD requires less training data and generalizes well to unseen classes and models. The authors demonstrate the effectiveness of HGD in the NIPS 2017 adversarial defense competition, where their solution won first place and outperformed other models significantly. The paper also discusses the transferability of HGD to different models and classes, and its role as an anti-adversarial transformer.The paper introduces a novel defense mechanism against adversarial attacks in image classification, called High-Level Representation Guided Denoiser (HGD). Traditional denoising methods suffer from the error amplification effect, where small adversarial perturbations are amplified to large errors in the model's output. HGD addresses this issue by using a loss function that measures the difference between the target model's outputs for clean and denoised images. This approach is more robust to both white-box and black-box adversarial attacks compared to ensemble adversarial training, which is the current state-of-the-art method. HGD requires less training data and generalizes well to unseen classes and models. The authors demonstrate the effectiveness of HGD in the NIPS 2017 adversarial defense competition, where their solution won first place and outperformed other models significantly. The paper also discusses the transferability of HGD to different models and classes, and its role as an anti-adversarial transformer.