mlsanityData to 7 Oct 2026

Exposing and Defending the Achilles' Heel of Video Mixture-of-Experts

arXiv 2602.01369 ↗First posted 1 Feb 2026cs.CVcited by 3

On the map: Mixture-of-experts → Vision mixture-of-experts

Abstract (arXiv)

Mixture-of-Experts (MoE) has demonstrated strong performance in video understanding tasks, yet its adversarial robustness remains underexplored. Existing attack methods often treat MoE as a unified architecture, overlooking the independent and collaborative weaknesses of key components such as routers and expert modules. To fill this gap, we propose Temporal Lipschitz-Guided Attacks (TLGA) to thoroughly investigate component-level vulnerabilities in video MoE models. We first design attacks on the router, revealing its independent weaknesses. Building on this, we introduce Joint Temporal Lipschitz-Guided Attacks (J-TLGA), which collaboratively perturb both routers and experts. This joint attack significantly amplifies adversarial effects and exposes the Achilles' Heel (collaborative weaknesses) of the MoE architecture. Based on these insights, we further propose Joint Temporal Lipschitz Adversarial Training (J-TLAT). J-TLAT performs joint training to further defend against collaborative weaknesses, enhancing component-wise robustness. Our framework is plug-and-play and reduces inference cost by more than 60% compared with dense models. It consistently enhances adversarial robustness across diverse datasets and architectures, effectively mitigating both the independent and collaborative weaknesses of MoE.

Key facts. Cited by 3 papers in the map (data through 7 Oct 2026). Placed in Vision mixture-of-experts, one of 12 directions in Mixture-of-experts. Builds on 46 of its 65 references that are in the map. First posted on arXiv 1 Feb 2026.

Builds on

46 of 65 references.

Closest ancestors · 8

Same direction, newest first

  1. 2025Heterogeneous Mixture of Experts for Remote Sensing Image Super-Resolution12 Feb
  2. 2025Optimizing Robustness and Accuracy in Mixture of Experts: A Dual-Model Approach5 Feb
  3. 20253D-MoE: Towards Spatial Intelligence with Mixture-of-Experts for 3D Reasoning and Action Generation28 Jan
  4. 2024Mixture of Experts in Image Classification: What's the Sweet Spot?27 Nov
  5. 2024Mixture of Nested Experts: Adaptive Processing of Visual Tokens29 Jul
  6. 2024Video Relationship Detection Using Mixture of Experts6 Mar
  7. 2024Enhancing the "Immunity" of Mixture-of-Experts Networks for Adversarial Defense29 Feb
  8. 2023Robust Mixture-of-Expert Training for Convolutional Neural Networks19 Aug

Other references · 38

  1. 2014Explaining and Harnessing Adversarial Examplescited by 7,287
  2. 2017Towards Deep Learning Models Resistant to Adversarial Attackscited by 6,326
  3. 2017Outrageously Large Neural Networks: The Sparsely-Gated Mixture-of-Experts Layercited by 3,413
  4. 2021Switch Transformers: Scaling to Trillion Parameter Models with Simple and Efficient Sparsitycited by 3,014
  5. 2018SlowFast Networks for Video Recognitioncited by 1,862
  6. 2020Reliable evaluation of adversarial robustness with an ensemble of diverse parameter-free attackscited by 1,442
  7. 2018TSM: Temporal Shift Module for Efficient Video Understandingcited by 852
  8. 2017Ensemble Adversarial Training: Attacks and Defensescited by 833
  9. 2020Fast is better than free: Revisiting adversarial trainingcited by 723
  10. 2017Can Spatiotemporal 3D CNNs Retrace the History of 2D CNNs and ImageNet?cited by 551
  11. 2017Mitigating Adversarial Effects Through Randomizationcited by 382
  12. 2020Training robust neural networks using Lipschitz boundscited by 98
26 more
  1. 2018Sparse Adversarial Perturbations for Videoscited by 56
  2. 2024VideoLLM-MoD: Efficient Video-Language Streaming with Mixture-of-Depths Vision Computationcited by 55
  3. 2024Multi-Modal Data-Efficient 3D Scene Understanding for Autonomous Drivingcited by 53
  4. 2022Rethinking Lipschitz Neural Networks and Certified Robustness: A Boolean Function Perspectivecited by 48
  5. 2019Heuristic Black-box Adversarial Attacks on Video Recognition Modelscited by 37
  6. 2025Stochastic Self-Guidance for Training-Free Enhancement of Diffusion Modelscited by 35
  7. 2022Isometric 3D Adversarial Examples in the Physical Worldcited by 25
  8. 2021Boosting the Transferability of Video Adversarial Examples via Temporal Translationcited by 21
  9. 2025DAST: Context-Aware Compression in LLMs via Dynamic Allocation of Soft Tokenscited by 21
  10. 2023Efficient Robustness Assessment via Adversarial Spatial-Temporal Focus on Videoscited by 20
  11. 2025Taming Preference Mode Collapse via Directional Decoupling Alignment in Diffusion Reinforcement Learningcited by 18
  12. 2023Temporal Coherent Test-Time Optimization for Robust Video Classificationcited by 15
  13. 2023Boosting Black-box Attack to Deep Neural Networks with Conditional Diffusion Modelscited by 12
  14. 2022Analysis and Extensions of Adversarial Training for Video Classificationcited by 11
  15. 2025ZeroTuning: Unlocking the Initial Token's Power to Enhance Large Language Models Without Trainingcited by 11
  16. 2024Efficient Black-box Adversarial Attacks via Bayesian Optimization Guided by a Function Priorcited by 10
  17. 2025Fast Adversarial Training with Weak-to-Strong Spatial-Temporal Consistency in the Frequency Domain on Videoscited by 9
  18. 2025Rethinking Machine Unlearning in Image Generation Modelscited by 9
  19. 2025Read Before You Think: Mitigating LLM Comprehension Failures with Step-by-Step Readingcited by 8
  20. 2025RunawayEvil: Jailbreaking the Image-to-Video Generative Modelscited by 7
  21. 2020Overcomplete Representations Against Adversarial Videoscited by 6
  22. 2023AFLOW: Developing Adversarial Examples under Extremely Noise-limited Settingscited by 4
  23. 2025Exploring Adversarial Transferability between Kolmogorov-arnold Networkscited by 4
  24. 2025An Effective End-to-End Solution for Multimodal Action Recognitioncited by 3
  25. 2025Anti-Aesthetics: Protecting Facial Privacy against Customized Text-to-Image Synthesiscited by 2
  26. 2025Beyond Detection: A Comprehensive Benchmark and Study on Representation Learning for Fine-Grained Webshell Family Classificationcited by 1

Built on by

Cited by year (2026 to 7 Oct)

Citing papers per year: 2026: 3.

3 papers.

  1. 2026LUVE : Latent-Cascaded Ultra-High-Resolution Video Generation with Dual Frequency Expertscited by 13
  2. 2026From Zero to Detail: A Progressive Spectral Decoupling Paradigm for UHD Image Restoration with New Benchmarkcited by 6
  3. 2026Hierarchical Anti-Aesthetics: Protecting Facial Privacy against Customized Diffusion Modelscited by 0

Similar, not cited

Close in wording, no citation either way

  1. 2024Towards Adversarial Robustness of Model-Level Mixture-of-Experts Architectures for Semantic Segmentation
  2. 2022On the Adversarial Robustness of Mixture of Experts
  3. 2026SEAL: Reinforcing Global Safety in Mixture-of-Experts through Shared Expert ALignment
  4. 2025Robust Experts: the Effect of Adversarial Training on CNNs with Sparse Mixture-of-Experts Layers
  5. 2026L2R: Low-Rank and Lipschitz-Controlled Routing for Mixture-of-Experts
  6. 2023Revisiting Single-gated Mixtures of Experts
  7. 2026RoME: Robust Mixture of Low-Rank Experts against Multiple Adversarial Perturbations
  8. 2025Backdoor or Manipulation? Graph Mixture of Experts Can Defend Against Various Graph Adversarial Attacks