A Physics-Inspired SU(2) Interferometric Pipeline for Anomalous Gravitational-Wave Detection

The detection of anomalous gravitational-wave (GW) signals, particularly those not well-described by existing theoretical templates, represents a significant challenge and a frontier in multi-messenger astronomy. Conventional deep learning approaches, while powerful, often function as data-intensive "black boxes" that lack physical interpretability and struggle with out-of-distribution events. This paper presents a fundamentally different paradigm: a lightweight, physics-inspired computational pipeline designed for the robust detection of anomalous GW signals. Our model explicitly avoids convolutional or transformer architectures, instead simulating the physical processes of optical interferometry and quantum-like dynamics. The core of the architecture consists of three physics-based stages: (1) an SU(2) interferometric mixer that coherently combines the two LIGO detector channels, transforming relative phase information into amplitude contrast; (2) a 1D temporal "4f" correlator that applies a low-parameter, physically-constrained filter in the frequency domain; and (3) a lightweight diffusion-decay mechanism that accumulates signal evidence over time. The system processes raw time-series data and extracts a small set of physically meaningful features, which are fed into a minimal linear classifier to produce a final anomaly score. On the A3D3 challenge dataset, our model achieves exceptional performance, reaching a True Positive Rate (TPR) of 0.9941 while maintaining a True Negative Rate (TNR) of nearly 1.0 on a synthetically calibrated hold-out set. Furthermore, the pipeline demonstrates extreme computational efficiency, processing over 12,500 samples per second on a single CPU core with a peak memory footprint of only 0.64 GB. This work demonstrates that by embedding physical principles directly into the model architecture, it is possible to achieve state-of-the-art detection performance, interpretability, and computational efficiency, offering a promising alternative for real-time anomaly detection in high-throughput scientific data streams.