Scaling Up Lensing Detection with Machine Learning

July 26, 2025 • Gravitational Waves • Machine Learning

Identifying strongly lensed gravitational‑wave events is both scientifically important and computationally challenging. Traditional Bayesian methods require pairwise parameter estimation for millions of signal pairs, which is infeasible.

📌 The ML Pipeline

• Generate Q‑transform spectrograms (time–frequency maps)
• Generate Bayestar skymaps for rapid localization
• Use a CNN (DenseNet) for signal feature extraction
• Combine outputs with XGBoost classifiers to rank lensing candidates

This hybrid model filters millions of candidate pairs in seconds while maintaining Bayesian-level accuracy.

🎯 Visual Illustration

Figure: Example of Q‑transform input images—a lensed pair (top row) and an unlensed pair (bottom row).

🔧 Results & Impact

The pipeline was validated on simulated non-spinning binary black hole signals with Gaussian noise. It saves orders of magnitude of computation—feasible for next-generation detectors like LISA and Einstein Telescope.

Stay tuned for open source code and live deployment!