mlsanityData to 7 Oct 2026

Certified Robustness under Heterogeneous Perturbations via Hybrid Randomized Smoothing

arXiv 2605.12876 ↗First posted 13 May 2026cs.LGcited by 0

Not on the map. Why?

Abstract (arXiv)

Randomized smoothing provides strong, model-agnostic robustness certificates, but existing guarantees are limited to single modalities, treating continuous and discrete inputs in isolation. This limitation becomes critical in multimodal models, where decisions depend on cross-modal semantics and adversaries can jointly perturb heterogeneous inputs, rendering unimodal certificates insufficient. We introduce a unified randomized smoothing framework for mixed discrete--continuous inputs based on an analytically tractable Neyman--Pearson formulation of the joint worst-case problem. By analyzing the joint likelihood ordering induced by factorized discrete and continuous noise, our approach yields a closed-form, one-dimensional certificate that strictly generalizes both Gaussian (image-only) and discrete (text-only) randomized smoothing. We validate the framework on multimodal safety filtering, providing, to our knowledge, the first model-agnostic Neyman--Pearson certificate for joint discrete-token and continuous-image perturbations in interaction-dependent text--image safety filtering.

Key facts. No citing papers in the map yet (data through 7 Oct 2026). Builds on 43 of its 58 references that are in the map. First posted on arXiv 13 May 2026.

Builds on

43 of 58 references.

References · 43

  1. 2021Masked Autoencoders Are Scalable Vision Learnerscited by 7,339
  2. 2017Towards Deep Learning Models Resistant to Adversarial Attackscited by 6,326
  3. 2023Universal and Transferable Adversarial Attacks on Aligned Language Modelscited by 2,914
  4. 2019Certified Adversarial Robustness via Randomized Smoothingcited by 1,324
  5. 2019Is BERT Really Robust? A Strong Baseline for Natural Language Attack on Text Classification and Entailmentcited by 663
  6. 2020The Hateful Memes Challenge: Detecting Hate Speech in Multimodal Memescited by 485
  7. 2018Certified Robustness to Adversarial Examples with Differential Privacycited by 462
  8. 2019Provably Robust Deep Learning via Adversarially Trained Smoothed Classifierscited by 328
  9. 2024StruQ: Defending Against Prompt Injection with Structured Queriescited by 325
  10. 2023Are aligned neural networks adversarially aligned?cited by 305
  11. 2023Low-Resource Languages Jailbreak GPT-4cited by 290
  12. 2023A Survey on Multimodal Large Language Models for Autonomous Drivingcited by 288
31 more
  1. 2025Defeating Prompt Injections by Designcited by 256
  2. 2023MM-SafetyBench: A Benchmark for Safety Evaluation of Multimodal Large Language Modelscited by 255
  3. 2023Certifying LLM Safety against Adversarial Promptingcited by 248
  4. 2019Certified Robustness to Adversarial Word Substitutionscited by 164
  5. 2022Concrete Score Matching: Generalized Score Matching for Discrete Datacited by 144
  6. 2022Towards Adversarial Attack on Vision-Language Pre-training Modelscited by 142
  7. 2022(Certified!!) Adversarial Robustness for Free!cited by 130
  8. 2025Adaptive Attacks Break Defenses Against Indirect Prompt Injection Attacks on LLM Agentscited by 112
  9. 2023Set-level Guidance Attack: Boosting Adversarial Transferability of Vision-Language Pre-training Modelscited by 106
  10. 2025Vision-Language-Action Models for Robotics: A Review Towards Real-World Applicationscited by 82
  11. 2020SAFER: A Structure-free Approach for Certified Robustness to Adversarial Word Substitutionscited by 72
  12. 2023VLATTACK: Multimodal Adversarial Attacks on Vision-Language Tasks via Pre-trained Modelscited by 66
  13. 2019Tight Certificates of Adversarial Robustness for Randomly Smoothed Classifierscited by 65
  14. 2019Robustness Certificates for Sparse Adversarial Attacks by Randomized Ablationcited by 59
  15. 2021Certified Robustness to Text Adversarial Attacks by Randomized [MASK]cited by 57
  16. 2024LlavaGuard: An Open VLM-based Framework for Safeguarding Vision Datasets and Modelscited by 49
  17. 2021Improved, Deterministic Smoothing for L_1 Certified Robustnesscited by 39
  18. 2025Defending Against Prompt Injection With a Few DefensiveTokenscited by 26
  19. 2024Your Diffusion Model is Secretly a Certifiably Robust Classifiercited by 24
  20. 2024Sports-Traj: A Unified Trajectory Generation Model for Multi-Agent Movement in Sportscited by 19
  21. 2025Membership Inference Attacks on Large-Scale Models: A Surveycited by 17
  22. 2024Adversarial Attacks to Multi-Modal Modelscited by 12
  23. 2025Towards the Worst-case Robustness of Large Language Modelscited by 7
  24. 2023The Lipschitz-Variance-Margin Tradeoff for Enhanced Randomized Smoothingcited by 6
  25. 2023Fast Certification of Vision-Language Models Using Incremental Randomized Smoothingcited by 4
  26. 2024MMCert: Provable Defense against Adversarial Attacks to Multi-modal Modelscited by 4
  27. 2024Prompt Injection Attacks in Defended Systemscited by 4
  28. 2024Achieving Domain-Independent Certified Robustness via Knowledge Continuitycited by 3
  29. 2024COMMIT: Certifying Robustness of Multi-Sensor Fusion Systems against Semantic Attackscited by 2
  30. 2025Bridging the Theoretical Gap in Randomized Smoothingcited by 2
  31. 2025Randomized Smoothing Meets Vision-Language Modelscited by 1

Built on by

Not cited yet.