The recent advancements in anomaly detection across various domains, including multi-class anomaly detection, time series anomaly detection, and industrial anomaly detection, have shown significant progress through innovative methodologies. A notable trend is the integration of contrastive learning and distillation techniques to enhance model performance and adaptability. In multi-class anomaly detection, the incorporation of class-aware contrastive learning has addressed issues like catastrophic forgetting and inter-class confusion, leading to superior performance. Time series anomaly detection has seen a shift towards faster, patch-based methods leveraging broad learning systems, offering a balance between efficiency and accuracy. Industrial anomaly detection has benefited from recalibrated attention mechanisms and multimodal approaches, which focus on decomposing and refining attention to better detect subtle defects. These developments collectively indicate a move towards more efficient, accurate, and scalable anomaly detection models, with a strong emphasis on leveraging diverse data modalities and learning techniques to overcome traditional limitations.