Share your thoughts, 1 month free Claude Pro on usSee more
WorkDL logo mark

Halt Fast! Early Stopping for Certified Robustness

About

Randomized Smoothing (RS) provides rigorous robustness guarantees for neural networks without architectural constraints, yet its adoption is limited by extreme computational costs. Standard RS requires tens of thousands of model evaluations per input and forces practitioners to commit to fixed sample sizes a priori. In this work, we present a novel meta-learning framework for anytime-valid certified robustness that adaptively deploys computational resources. By using a lightweight meta-learner to predict image-specific priors for a sequential E-process, we achieve a 20-fold reduction in sample complexity compared to traditional methods while maintaining rigorous statistical guarantees. Beyond raw efficiency, we demonstrate how anytime-validity enables adaptively allocating compute based upon application-specific risk thresholds, a form of resource triage impossible under classic certification frameworks. That this is achievable while also providing similar certification performance demonstrates that our approach provides a pathway for real-time, safety-critical certification deployments.

Andrew C. Cullen, Paul Montague, Benjamin I.P. Rubinstein• 2026

Related benchmarks

TaskDatasetResultRank
Certified RobustnessMNIST
Certified Accuracy (R=0.00)99.2
24
Image ClassificationMNIST
Accuracy99.2
15
Image ClassificationImageNet
Accuracy72.6
15
Image ClassificationCIFAR-10
Accuracy74
15
Certified RobustnessImageNet--
8
Certified RobustnessCIFAR-10
Certified Accuracy78.8
6
Operational TriageCIFAR-10 (test)
Average Samples1.45e+3
4
Showing 7 of 7 rows

Other info

Follow for update