Benchmarking Resilience: Asymmetric Latent Purification Inspired by Generative Diffusion Bottlenecks

Deep learning classifiers exhibit susceptibility towards iterative adversarial perturbations, often under high-fidelity attacks experiencing total categorical collapse. To address this, we introduce the Asymmetric Latent Purifier (ALP), a novel structural defence mechanism inspired by the stochastic information bottlenecks of the 2026 Unified Latents (UL) generative framework, Unlike Traditional deterministic autoencoders, ALP incorporates an adaptive, non-differentiable Gaussian noise layer within a 64-channel latent manifold to disrupt adversarial gradient flows. Empirically validated on CIFAR-10 dataset using an Apple M4 8-core GPU architecture. While the unprotected baseline experiences a total categorical collapse ( 0.00% accuracy) under a 7-step iterative PGD attack, our 20-sample adaptive ensemble approach achieves a robust accuracy of 32.06% (SD=1.94%)( averaged over 5 trials ) while ensuring a high-fidelity reconstruction of 25.68 dB. Operating a total system latency of 13.86ms, offers a promising path towards real-time flexibility for complex RGB varieties. Furthermore, with a single-sample inference latency of 1.25 ms, ALP represents a 100x to 1000x speedup over iterative diffusion-based purifiers, enabling real-time adversarial immunity in safety-critical systems.

Benchmarking, Asymmetric , Purification Inspired

1. J. Heek et al., "Unified Latents (UL): How to train your latents," arXiv:2602.17270, 2026. [Google Scholar] [Crossref]

2. M. Jensen et al., "Adaptive Compression and Quantization Techniques for Robust and Scalable Generative Diffusion Networks," TechRxiv:10.36227/techrxiv.175693734.42139855, 2025. [Google Scholar] [Crossref]

3. J. Wu et al., "Ptq4dit: Post-training quantization for diffusion transformers," arXiv:2405.16005, 2024. [Google Scholar] [Crossref]

4. X. Li et al., "Q-diffusion: Quantizing diffusion models," arXiv:2302.04304, ICCV, 2023.[5] J. Ho, A. Jain, and P. Abbeel, "Denoising diffusion probabilistic models," arXiv:2006.11239, NeurIPS, 2020.[6] Y. Song et al., "Score-based generative modeling through stochastic differential equations," arXiv:2011.13456, ICLR, 2021. [Google Scholar] [Crossref]

5. C. Szegedy et al., "Intriguing properties of neural networks," arXiv:1312.6199, ICLR, 2014. [Google Scholar] [Crossref]

6. I. J. Goodfellow, J. Shlens, and C. Szegedy, "Explaining and harnessing adversarial examples," arXiv:1412.6572, ICLR, 2015. [Google Scholar] [Crossref]

7. A. Shafahi et al., "Are adversarial examples inevitable?," arXiv:1809.02104, ICLR, 2019. [Google Scholar] [Crossref]

8. A. Ilyas et al., "Adversarial examples are not bugs, they are features," arXiv:1905.02175, NeurIPS, 2019. [Google Scholar] [Crossref]

9. A. Madry et al., "Towards deep learning models resistant to adversarial attacks," arXiv:1706.06083, ICLR, 2018. [Google Scholar] [Crossref]

10. N. Carlini and D. Wagner, "Towards evaluating the robustness of neural networks," arXiv:1608.04644, IEEE S&P, 2017. [Google Scholar] [Crossref]

11. W. Brendel, J. Rauber, and M. Bethge, "Decision-based adversarial attacks," arXiv:1712.04248, ICLR, 2018. [Google Scholar] [Crossref]

12. N. Papernot et al., "Distillation as a defense to adversarial perturbations against deep neural networks," arXiv:1511.04508, IEEE S&P, 2016. [Google Scholar] [Crossref]

13. A. Kurakin, I. Goodfellow, and S. Bengio, "Adversarial examples in the physical world," arXiv:1607.02533, ICLR Workshop, 2017. [Google Scholar] [Crossref]

14. T. B. Brown et al., "Unrestricted adversarial examples," arXiv:1809.08352, 2018. [Google Scholar] [Crossref]

15. A. Athalye, N. Carlini, and D. Wagner, "Obfuscated gradients give a false sense of security," arXiv:1802.00420, ICML, 2018. [Google Scholar] [Crossref]

16. F. Croce and M. Hein, "Reliable evaluation of adversarial robustness with an ensemble of diverse parameter-free attacks," arXiv:2003.01690, ICML, 2020. [Google Scholar] [Crossref]

17. D. Tsipras et al., "Robustness may be at odds with accuracy," arXiv:1805.12152, ICLR, 2019. [Google Scholar] [Crossref]

18. J. Rauber, W. Brendel, and M. Bethge, "Foolbox: A python toolbox to benchmark robustness," arXiv:1707.04131, 2017. [Google Scholar] [Crossref]

19. F. Tramèr et al., "Ensemble adversarial training: Attacks and defenses," arXiv:1705.07204, ICLR, 2018. [Google Scholar] [Crossref]

20. Y. Dong et al., "Benchmarking adversarial robustness on image classification," CVPR, 2020. [Google Scholar] [Crossref]

21. Z. Cui et al., "On the robustness of large multimodal models against image adversarial attacks," arXiv:2312.03777, CVPR, 2024. [Google Scholar] [Crossref]

22. A. Rusiecki and Y. T-Menard, "A comprehensive study on robustness of image classification models," arXiv:2302.14301, IJCV, 2021. [Google Scholar] [Crossref]

23. A. Al-hajjar and A. Al-khayer, "Adversarial attacks on image classification models: Analysis and defense," arXiv:2312.16880, 2023. [Google Scholar] [Crossref]

24. D. Meng and H. Chen, "MagNet: a two-pronged defense against adversarial examples," arXiv:1705.09064, CCS, 2017. [Google Scholar] [Crossref]

25. F. Liao et al., "Defense against adversarial attacks using high-level representation guided denoiser," arXiv:1712.02976, CVPR, 2018. [Google Scholar] [Crossref]

26. N. Carlini and D. Wagner, "Magnet and efficient defenses against adversarial attacks are not robust," arXiv:1711.08478, 2017. [Google Scholar] [Crossref]

27. W. Nie, B. Guo, Y. Huang, C. Xiao, A. Vahdat, and A. Anandkumar, "Diffusion models for adversarial purification," arXiv:2205.07460, ICML, 2022. [Google Scholar] [Crossref]

28. C. Xiao, Z. Zhong, D. Zheng, L. Yuan, and G. Huang, "DensePure: Understanding diffusion purification towards robust classifiers," arXiv:2211.00322, ICLR, 2023. [Google Scholar] [Crossref]

29. J. Wang, Z. Lyu, D. Lin, B. Dai, and H. Xu, "Guided diffusion model for adversarial purification," arXiv:2205.14969, 2023.[32] J. M. Cohen, E. Rosenfeld, and J. Z. Kolter, "Certified adversarial robustness via randomized smoothing," arXiv:1902.02918, ICML, 2019. [Google Scholar] [Crossref]

30. C. J. Simon-Gabriel and H. Meuel, "Evaluating adversarial robustness on document image classification," arXiv:2304.12486, ICDAR, 2023. [Google Scholar] [Crossref]

31. M. Al-braithen et al., "Evaluating the robustness of deep learning models against adversarial attacks," Big Data Cogn. Comput., 2023. [Google Scholar] [Crossref]

• What the Desert Fathers Teach Data Scientists: Ancient Ascetic Principles for Ethical Machine-Learning Practice
• Comparative Analysis of Some Machine Learning Algorithms for the Classification of Ransomware
• Comparative Performance Analysis of Some Priority Queue Variants in Dijkstra’s Algorithm
• Transfer Learning in Detecting E-Assessment Malpractice from a Proctored Video Recordings.
• Dual-Modal Detection of Parkinson’s Disease: A Clinical Framework and Deep Learning Approach Using NeuroParkNet