arXiv Analytics

Sign in

arXiv:2302.01332 [cs.LG]AbstractReferencesReviewsResources

Bayesian Metric Learning for Uncertainty Quantification in Image Retrieval

Frederik Warburg, Marco Miani, Silas Brack, Soren Hauberg

Published 2023-02-02Version 1

We propose the first Bayesian encoder for metric learning. Rather than relying on neural amortization as done in prior works, we learn a distribution over the network weights with the Laplace Approximation. We actualize this by first proving that the contrastive loss is a valid log-posterior. We then propose three methods that ensure a positive definite Hessian. Lastly, we present a novel decomposition of the Generalized Gauss-Newton approximation. Empirically, we show that our Laplacian Metric Learner (LAM) estimates well-calibrated uncertainties, reliably detects out-of-distribution examples, and yields state-of-the-art predictive performance.

Comments: Code:
Categories: cs.LG, cs.CV
Related articles: Most relevant | Search more
arXiv:2310.09338 [cs.LG] (Published 2023-10-13)
Uncertainty Quantification using Generative Approach
arXiv:2210.00074 [cs.LG] (Published 2022-09-30)
Leveraging Industry 4.0 -- Deep Learning, Surrogate Model and Transfer Learning with Uncertainty Quantification Incorporated into Digital Twin for Nuclear System
M. Rahman et al.
arXiv:2403.10642 [cs.LG] (Published 2024-03-15)
Using Uncertainty Quantification to Characterize and Improve Out-of-Domain Learning for PDEs