Advancements in Anomaly Detection and Generalized Category Discovery

The field of anomaly detection is moving towards more generalizable and adaptable models, with a focus on overcoming the limitations of specialized methods. Researchers are exploring the use of mixture-of-experts architectures and adaptive thresholding frameworks to improve the detection of anomalies in various domains, including visual, video, and time series data. Additionally, there is a growing interest in developing frameworks that can learn from limited labeled data and adapt to changing environments. Noteworthy papers in this area include AnomalyMoE, which proposes a novel anomaly detection framework based on a mixture-of-experts architecture, and GS-MoE, which introduces a Gaussian splatting-guided mixture of experts approach for weakly-supervised video anomaly detection. Other notable works include Segmented Confidence Sequences and Multi-Scale Adaptive Confidence Segments for anomaly detection in nonstationary time series, ALFred, an active learning framework for real-world semi-supervised anomaly detection, and ConGCD, a consensus-aware paradigm for generalized category discovery.

Sources

AnomalyMoE: Towards a Language-free Generalist Model for Unified Visual Anomaly Detection

Mixture of Experts Guided by Gaussian Splatters Matters: A new Approach to Weakly-Supervised Video Anomaly Detection

Segmented Confidence Sequences and Multi-Scale Adaptive Confidence Segments for Anomaly Detection in Nonstationary Time Series

ALFred: An Active Learning Framework for Real-world Semi-supervised Anomaly Detection with Adaptive Thresholds

Dissecting Generalized Category Discovery: Multiplex Consensus under Self-Deconstruction

Built with on top of