JOHN VADEN

Greetings. I am John Vaden, a machine learning astrophysicist specializing in self-supervised representation learning to disentangle faint noise artifacts from gravitational wave (GW) signals. With a Ph.D. in Computational Astrophysics (MIT, 2024) and a Senior Research Scientist role at LIGO Laboratory, I pioneer adaptive noise-mitigation frameworks that enhance the sensitivity of interferometric detectors like LIGO, Virgo, and the upcoming Einstein Telescope. My work bridges deep learning theory and observational cosmology to address one of the most pressing challenges in GW astronomy: detecting sub-threshold signals buried in non-Gaussian, non-stationary noise.

Methodological Innovations

1. Contrastive Embedding for Noise Invariance

• Challenge: Traditional supervised models fail to generalize across detector upgrades (e.g., LIGO A+/AdV+) and novel noise sources (e.g., quantum squeezing artifacts).

• Breakthrough: Developed GW-SimCLR, a self-supervised framework that:

  • Learns noise-invariant embeddings by contrasting time-frequency representations of real detector data and synthetic glitches.

  • Achieves 92% precision in identifying microlensing-correlated noise (e.g., scattered light events) without labeled training data (Nature Machine Intelligence, 2025).

  • Open-sourced as part of the GWLab toolkit (GitHub Stars: 2,800+).

2. Multi-Detector Masked Autoencoding

• Architecture: GW-MAE reconstructs masked segments of cross-correlated strain data from LIGO-Hanford, LIGO-Livingston, and Virgo:

  • Trained on O4/O5 observing run data (2 PB), capturing transient noise (e.g., cosmic ray showers, violin-mode resonances).

  • Enables few-shot adaptation to KAGRA and future detectors via latent space alignment.

• Impact: Reduced false alarm rates by 40% in sub-threshold binary neutron star searches.

3. Uncertainty-Aware Anomaly Scoring

• Toolkit: NoiseBERT combines:

  • Probabilistic Attention: Quantifies epistemic uncertainty in noise classification.

  • Causal Discovery: Identifies root causes of non-astrophysical triggers (e.g., power grid fluctuations at LIGO sites).

• Deployment: Integrated into the GWOSC alert pipeline, prioritizing signals for EM follow-up (e.g., BlackGEM, Rubin Observatory).

Landmark Contributions

1. Unveiling "Silent" Glitches in LIGO O4 Data

• Discovery: Identified 23 previously undetected noise lineages (e.g., thermally driven suspension fiber vibrations) using topological data analysis (TDA) on GW-SimCLR embeddings.

• Impact: Informed hardware stabilization protocols, improving detector duty cycles by 15%.

2. Enabling High-Redshift GW Detection

• Problem: Quantum noise dominates at low frequencies (<30 Hz), obscuring primordial black hole mergers.

• Solution: Trained GW-MAE on simulated next-generation detector data (Einstein Telescope, Cosmic Explorer).

• Result: Boosted SNR for z > 10 mergers by 3×, enabling tests of early universe cosmology.

3. Citizen Science Integration

• Project: Launched GravitySpy 2.0, a crowdsourcing platform where volunteers label noise types using NoiseBERT-guided interfaces.

• Scale: Annotated 500,000+ glitches for public ML challenges, accelerating noise model retraining.

Technical and Collaborative Impact

1. Open Frameworks for Reproducibility

• Released GW-Torch, a PyTorch library with:

  • Pre-trained self-supervised models (GW-SimCLR, GW-MAE).

  • Synthetic noise generators calibrated to LIGO/Virgo logbooks.

• Adoption: Used by 300+ research groups in the LVC (LIGO-Virgo Collaboration).

2. Quantum-Inspired Optimization

• Collaboration with Google Quantum AI:

  • Implemented quantum kernel alignment to enhance contrastive learning efficiency.

  • Achieved 20% faster convergence on 50-qubit quantum simulators.

3. Ethics of Noise Democracy

• Authored LVC Policy on Noise Transparency (2025):

  • Mandates real-time noise metadata sharing to reduce geographic bias (e.g., Northern vs. Southern hemisphere detectors).

  • Establishes ethical guidelines for AI-generated noise injections.

Future Directions

1. Multi-Messenger Noise Cancellation:
   Correlate GW data with neutrino (IceCube) and cosmic ray (Pierre Auger) observatories to isolate atmospheric perturbations.

2. Edge AI for Real-Time Denoising:
   Deploy FPGA-accelerated GW-SimCLR at detector sites for sub-millisecond glitch vetoing.

3. Exascale Self-Supervision:
   Scale training to Einstein Telescope’s exabyte-scale data using distributed contrastive learning on DOE supercomputers.