This research introduces a novel One-to-Normal framework that significantly improves few-shot anomaly detection across security, factory, and medical applications.

• Transforms anomalous patterns into normal ones using a personalized approach
• Works with as few as 1-16 normal samples (few-shot learning)
• Outperforms existing methods by focusing on personalized feature transformation
• Demonstrates strong performance across multiple domains including security monitoring

For security applications, this approach enables more accurate threat detection with minimal examples, allowing systems to quickly adapt to new environments without extensive training data.

One-to-Normal: Anomaly Personalization for Few-shot Anomaly Detection