Self-Supervised Learning for Anomaly Detection in Large-Scale Streaming Data

Priyaranjan Pattnayak

Abstract:

Anomaly detection in large-scale streaming data is crucial for applications such as cybersecurity, fraud detection, and industrial monitoring. Traditional supervised learning approaches often require large labeled datasets, which are expensive and infeasible for dynamically evolving data streams. Self-supervised learning (SSL) has emerged as a powerful paradigm for unsupervised representation learning, enabling more effective anomaly detection without labeled data. In this paper, we propose a novel self-supervised anomaly detection framework that leverages contrastive learning and dynamic representation learning to detect anomalies in streaming data efficiently. Our approach adapts to evolving data distributions and outperforms traditional baselines in multiple real-world datasets.
Keywords: Self-Supervised Learning, Anomaly Detection, Streaming Data, Unsupervised Learning, Data Mining