← Back to projects
Detecting Gravitational Waves with CNNs
Built a convolutional neural network pipeline that detects gravitational waves from binary black holes (BBH), binary neutron stars (BNS), and neutron star–black hole (NSBH) mergers using open LIGO GWOSC data. Deliverables included annotated notebooks, model explainability artifacts, and presentation materials for a graduate machine learning course.
Approach
- Curated training corpora from real and simulated LIGO GWOSC strain data, labeling BBH, BNS, and NSBH events to balance the dataset.
- Applied band-pass filtering, whitening, and windowing to boost signal-to-noise ratio prior to feeding tensors into the network.
- Architected a multi-branch CNN with residual connections to capture temporal patterns and reduce overfitting on noisy inputs.
- Implemented Grad-CAM plots to interpret salient regions and communicated findings to faculty and peers.
Highlights
- Achieved 90% confidence on the GW170817 neutron star merger and produced calibration curves that validated model reliability.
- Delivered reproducible notebooks with automated environment setup, enabling classmates to rerun experiments with minimal friction.
- Shared learnings on signal preprocessing and explainability in a lightning talk that informed future cohorts' project scoping.