🤖 AI Summary
This study investigates the reliability of latent representations learned by a denoising variational autoencoder (VAE-MoG) on the gravitational-wave event GW150914—specifically, whether the encoder’s output faithfully captures the true posterior structure despite high signal reconstruction fidelity. Method: We introduce Hamiltonian Monte Carlo (HMC) sampling from the exact posterior given clean input data and statistically compare these samples against encoder-inferred latent distributions. Contribution/Results: Experiments reveal substantial distributional mismatch between the encoder’s latent outputs and the ground-truth posterior, demonstrating that reconstruction accuracy alone overestimates latent-space trustworthiness. This work presents the first application of HMC-based posterior sampling to validate VAE latent spaces in gravitational-wave analysis. It exposes critical limitations of conventional evaluation paradigms and advocates posterior consistency—i.e., alignment between the generative model’s implied posterior and the true posterior—as a rigorous new evaluation criterion. The framework provides methodological grounding for interpretable latent-variable modeling in gravitational-wave data analysis.
📝 Abstract
In this work, we explore the latent space of a denoising variational autoencoder with a mixture-of-Gaussians prior (VAE-MoG), trained on gravitational wave data from event GW150914. To evaluate how well the model captures the underlying structure, we use Hamiltonian Monte Carlo (HMC) to draw posterior samples conditioned on clean inputs, and compare them to the encoder's outputs from noisy data. Although the model reconstructs signals accurately, statistical comparisons reveal a clear mismatch in the latent space. This shows that strong denoising performance doesn't necessarily mean the latent representations are reliable highlighting the importance of using posterior-based validation when evaluating generative models.