Research map / Adversarial robustness and security / Certified and empirical robustness
Biologically Inspired Mechanisms for Adversarial Robustness
Read the paperCheck your idea against 108,967 papers
AI-extracted summary. Written by AI from the paper's title only (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 retinal (non-uniform) fixation sampling instead of standard input | ResNet-18 · ImageNet clean | top-1 accuracy (%): 59.46 -> 62.90 | Helped |
| Use retinal fixations instead of standard ResNet | ResNet-20 · CIFAR10 clean | top-1 accuracy (%): 88.13 -> 88.88 | Helped |
| Use cortical (multi-scale) fixations instead of standard ResNet | ResNet-18 · ImageNet clean | top-1 accuracy (%): 59.46 -> 56.32 | Hurt |
| Add retinal and cortical fixation mechanisms | ResNet · small-epsilon PGD robustness, ImageNet10/100 | adversarial accuracy | Helped |
| Add retinal and cortical fixation mechanisms | ResNet · large-epsilon PGD robustness | adversarial accuracy | No clear effect |
| Replace non-uniform retinal sampling with uniform sub-sampling then up-sampling (ablation) | retinal fixations · ImageNet10 PGD robustness | adversarial accuracy | Hurt |
| Replace cortical multi-scale branches with a branched ensembling network (ablation) | cortical fixations · ImageNet10 PGD robustness | adversarial accuracy | Hurt |
| Replace cortical fixations with Gaussian blur on sub-samples only (ablation) | cortical fixations · ImageNet10 PGD robustness | adversarial accuracy | Hurt |
| Use only the largest-scale Gaussian-downsample branch instead of all scales (ablation) | cortical fixations · ImageNet10 PGD robustness | adversarial accuracy | Hurt |
| Pool across scale branches or use large dropout instead of concatenation (ablation) | cortical fixations · ImageNet10 targeted PGD | adversarial accuracy | Helped |
| Combine retinal sampling then cortical sampling (ablation) | retinal and cortical fixations · ImageNet10 targeted PGD | adversarial accuracy | No clear effect |
| Train cortical fixations model on bounding-box crops only (ablation) | cortical fixations · ImageNet10 targeted PGD | adversarial accuracy | No clear effect |
Closest papers by meaning
- Adversarially trained neural representations are already as robust as biological neural representations (ICML 2022 · similarity 0.92)
- A Combination of Noise and Bilateral Filters Achieve Supralinear and Scalable Adversarial Robustness in CNNs (CVPR 2026 · similarity 0.89)
- Towards robust vision by multi-task learning on monkey visual cortex (NeurIPS 2021 · similarity 0.89)
- Enhancing Adversarial Robustness in SNNs with Sparse Gradients (ICML 2024 · similarity 0.87)
- Adversarial Training Can Provably Improve Robustness: Theoretical Analysis of Feature Learning Process Under Structured Data (ICLR 2025 · similarity 0.87)
- Explicitly Modeling Subcortical Vision with a Neuro-Inspired Front-End Improves CNN Robustness (NeurIPS 2025 · similarity 0.86)
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.