Introduces Bayesian scattering as a mathematically grounded, non-learned baseline for image uncertainty quantification.
March 24, 2026
Original Paper
Bayesian Scattering: A Principled Baseline for Uncertainty on Image Data
arXiv · 2603.20908
The Takeaway
Provides a 'Bayesian Linear Regression' equivalent for image data, using fixed wavelet transforms to avoid the overfitting inherent in deep learning uncertainty methods. It offers a much-needed, interpretable benchmark for reliability in high-stakes domains like medical imaging.
From the abstract
Uncertainty quantification for image data is dominated by complex deep learning methods, yet the field lacks an interpretable, mathematically grounded baseline. We propose Bayesian scattering to fill this gap, serving as a first-step baseline akin to the role of Bayesian linear regression for tabular data. Our method couples the wavelet scattering transform-a deep, non-learned feature extractor-with a simple probabilistic head. Because scattering features are derived from geometric principles ra