AI Paper Atlas

Research map / Probabilistic and scientific modeling / Uncertainty and Gaussian processes

Frozen Priors, Fluid Forecasts: Prequential Uncertainty for Low-Data Deployment with Pretrained Generative Models

Fernando Ruiz-Mazo, Vikas K Garg

ICLR 2026 · Probabilistic and scientific modeling › Uncertainty and Gaussian processes · approach: Epistemic versus aleatoric decomposition

Read the paperCheck your idea against 137,217 papers

AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.

Measured effects reported in the paper

What was changedWhereMeasuredEffect
use DWS (Dirichlet-weighted shrinkage toward frozen GPT-2) instead of nonparametric bootstrapuncertainty intervals · GPT-2 WikiText-2, n0=20coverage@90%: 0.500 -> 0.900Improved
use DWS instead of Bayesian bootstrapuncertainty intervals · GPT-2 WikiText-2, n0=5coverage@90%: 0.000 -> 1.000Improved
use DWS at larger n0uncertainty intervals · GPT-2 WikiText-2, n0=100coverage@90%: 0.900 vs 0.900 (BB, JK)No clear effect
use martingale posterior instead of nonparametric bootstrapuncertainty intervals · Two Moons theta1, n0=5coverage@90%: 0.175 -> 0.825Improved
use martingale posterior instead of nonparametric bootstrap at large nuncertainty intervals · Two Moons theta1, n0=1000coverage@90%: 0.925 -> 0.825Hurt
increase real samples n0MP interval · Two Moons theta1interval width: 0.126 -> 0.017Improved
use DWS instead of bootstrapinterval width · GPT-2 WikiText-2, small nwidthHurt
use MP rather than nonparametric bootstrap on theta2uncertainty intervals · Two Moons theta2coverage@90%: n0=10: 0.975 vs 0.900; n0=20: 0.875 vs 0.900Mixed
fixed lambda (no decay) or non-predictable weights (ablation)forecast blend · theoryscalar coherenceHurt

Closest papers by meaning

Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.

Open scime draft

Get the next scime Dispatch and a monthly field update

Free. We send a confirmation link first; every email has a one-click unsubscribe. Privacy