Periodic model retraining can significantly improve the performance of anomaly detection models over time, but the choice of retraining technique (blind vs informed) and data (full-history vs sliding window) depends on the specific anomaly detection model.
AD-NEv++ is a neuroevolution-based framework that synergically combines subspace evolution, model evolution, and fine-tuning to optimize autoencoder architectures, including graph-based models, for multivariate anomaly detection. The framework supports a wide spectrum of neural layers, including attention, dense, and graph convolutional layers, and outperforms well-known deep learning architectures and neuroevolution-based approaches on benchmark datasets.
MambaAD, a novel framework that leverages the Mamba architecture, achieves state-of-the-art performance in multi-class unsupervised anomaly detection tasks while maintaining low model complexity.
Lifelong learning can provide significant advantages for anomaly detection models by enabling simultaneous adaptation and knowledge retention, leading to more robust and comprehensive models that can effectively handle dynamic environments.
LTAD combines anomaly detection by reconstruction and semantic anomaly detection to detect defects across multiple and long-tailed image classes, without relying on dataset class names. It learns pseudo-class names and uses a VAE-based data augmentation to address the long-tailed distribution of real-world applications.
RealNet introduces innovative anomaly detection methods with synthetic anomaly generation, feature selection, and reconstruction residuals.
주파수 관점에서 적은 샷 이상치 감지를 위한 이중 경로 주파수 판별기의 제안
Introducing the MINT-AD model for multi-class anomaly detection, leveraging class-aware query embeddings to mitigate inter-class interference.
Innovative self-supervised anomaly detection framework integrating synthetic anomaly generation.
提案されたDual-Path Frequency Discriminators(DFD)ネットワークは、周波数の観点から少数ショットの異常検出問題に取り組む革新的なアプローチを提供します。