AI Paper Atlas

Research map / Learning theory and algorithms / Prediction and decision losses

Invariance-Aware Randomized Smoothing Certificates

Jan Schuchardt, Stephan Günnemann

NeurIPS 2022 · Learning theory and algorithms › Prediction and decision losses · approach: Probabilistic calibration

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 the tight 2D rotation-invariance certificate instead of black-box smoothing2D rotation-invariant models · Adversarial scaling with σ=0.5, ||X||=0.01, p=0.8maximum certified ||Δ||₂: 0.4 -> 0.73Improved
Use rotation-aware gray-box certificates rather than black-box randomized smoothingRotation-invariant point-cloud classifiers · MNIST and ModelNet40, randomly perturbed inputs with ||Δ||=σ=0.1probabilistic certified accuracyImproved
Use the tight rotation certificate instead of the orbit-based certificateRotation-invariant point-cloud classifiers · MNIST and ModelNet40, perturbations with rotational componentsprobabilistic certified accuracyNo clear effect
Enforce translation invariance in addition to rotation invarianceEnsPointNet · MNIST point clouds, σ=0.2, adversarial scalingmaximum certifiable ||Δ||₂: ≈ 0.62 -> 0.8Improved
Use the tight certificate instead of the orbit-based certificateRotation-invariant classifiers · Point-cloud classification in practical smoothing regimescertified accuracyNo clear effect
Increase clean-data norm relative to smoothing standard deviationTight 2D rotation certificate · 2D rotation-invariance certificate parameter spacep_min improvement over black-box certificateHurt

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