Track papers across labs, topics, and trends.
Search papers, labs, and topics across Lattice.
This paper introduces Ribbon, a scalable method for approximating Dirichlet-reweighted bootstrap uncertainty that circumvents the computational costs associated with traditional Bayesian and bootstrap methods. By utilizing influence-function linearization around a single fitted model, Ribbon maintains the essential data-reweighting structure while significantly reducing the need for repeated model refitting. The method demonstrates competitive predictive performance and enhanced calibration across various benchmarks, including synthetic regression and MNIST classification, while ensuring efficiency in uncertainty quantification.
Ribbon achieves robust uncertainty quantification without the computational burden of repeated model retraining, making it a game-changer for high-dimensional predictive tasks.
Reliably quantifying predictive uncertainty is difficult for complex, high-dimensional, or misspecified models. Both fully Bayesian and bootstrap resampling methods provide principled uncertainty estimates but are often too expensive for modern machine-learning models because they require posterior sampling or repeated model refitting. We introduce Ribbon, a scalable approximation to Dirichlet-reweighted bootstrap uncertainty. Ribbon replaces repeated refitting with an influence-function linearization around a single fitted model, preserving the first-order data-reweighting structure of the Bayesian bootstrap while requiring only post-hoc linear algebra. Ribbon approximates the Bayesian-bootstrap or weighted-likelihood-bootstrap refitting target. With a general concentration parameter, Ribbon gives a calibrated Dirichlet-reweighting family whose uncertainty scale can be tuned on validation data. We show that Ribbon is asymptotically equivalent to a flat-prior Laplace approximation under correct likelihood specification and recovers the robust sandwich covariance under misspecification. Across synthetic regression, MNIST classification, and California Housing benchmarks, Ribbon provides competitive predictive performance and improved calibration in several settings while avoiding repeated model retraining.