Research map / Vision-language and multimodal / Cross-modal learning
Certified Robustness under Heterogeneous Perturbations via Hybrid Randomized Smoothing
Blaise Delattre, Hengyu WU, Paul Caillon, Wei Yang Bryan Lim, Yang Cao
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 changed | Where | Measured | Effect |
|---|---|---|---|
| use hybrid certificate instead of image-only RS | Hybrid RS · interaction-only suffix attack | image radius: 3.99 -> 3.76 | Hurt |
| use hybrid certificate instead of text-only RS | Hybrid RS · interaction-only suffix attack | text budget: 3.26 -> 3.07 | Hurt |
| use hybrid certificate instead of text-only RS | Hybrid RS · l0 text attack | mean certified l0 radius: 1.020 -> 0.33 | Hurt |
| add joint certification (unimodal certificates unsound) | Hybrid RS · interaction-only | soundness | Improved |
| increase corruption rate beta from 0.1 to 1.0 (ablation) | Hybrid RS · Hateful Memes | certified examples (%): 82.35 -> 41.18 | Hurt |
| increase corruption rate beta from 0.1 to 1.0 (ablation) | Hybrid RS · Hateful Memes | mean d_max: 2.29 -> 8.00 | Improved |
| increase Gaussian sigma from 0.5 to 1.0 (ablation) | Hybrid RS · Hateful Memes (eps=0.5, d=3) | certified accuracy (%): 38 -> 36 | Mixed |
| increase text budget d | Hybrid RS · Hateful Memes | certified accuracy | Hurt |
| use one-shot suffix certification with d_max=8 | Hybrid RS · certification cost | avg time and radius: 500 s -> 44 s; r 2.07 -> 1.55; d 1.48 -> 1.23 | Mixed |
| increase attack text budget from d* to d*+4 | adaptive attack · Hybrid RS | Safe%: 74.286 -> 90.214 | Improved |
| increase attack image budget from r* to 3r* | adaptive attack · Hybrid RS | Safe%: 74.286 -> 76.571 | No clear effect |
Closest papers by meaning
- Certified Robustness for Deep Equilibrium Models via Serialized Random Smoothing (NeurIPS 2024 · similarity 0.87)
- Incremental Randomized Smoothing Certification (ICLR 2024 · similarity 0.87)
- Mitigating the Curse of Dimensionality for Certified Robustness via Dual Randomized Smoothing (ICLR 2024 · similarity 0.86)
- Adaptive Randomized Smoothing: Certified Adversarial Robustness for Multi-Step Defences (NeurIPS 2024 · similarity 0.86)
- Certified Adversarial Robustness via Randomized $\alpha$-Smoothing for Regression Models (NeurIPS 2024 · similarity 0.85)
- Halt Fast! Early Stopping for Certified Robustness (NeurIPS 2026 · similarity 0.85)
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.
