Research map / Adversarial robustness and security / Attacks on text and graphs
Simulating a Primary Visual Cortex at the Front of CNNs Improves Robustness to Image Perturbations
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 |
|---|---|---|---|
| replace first block with fixed V1-like VOneBlock | ResNet50 · ImageNet perturbation benchmark | perturbation mean accuracy: 27.6±0.1 -> 45.6±0.2 | Helped |
| VOneBlock front-end | ResNet50 · ImageNet white box PGD | accuracy: 16.4±0.1 -> 51.1±0.4 | Helped |
| VOneBlock front-end | ResNet50 · ImageNet clean | accuracy: 75.6±0.1 -> 71.7±0.1 | Hurt |
| VOneBlock front-end | ResNet50 · ImageNet-C | accuracy: 38.8±0.3 -> 40.0±0.3 | Helped |
| add V1 stochasticity (ablation) | VOneResNet50 · ImageNet white box PGD | accuracy: -30.6±0.4 | Helped |
| add V1 stochasticity (ablation) | VOneResNet50 · ImageNet-C | accuracy: 1.5±0.5 | Hurt |
| add high SF Gabors (ablation) | VOneResNet50 · ImageNet white box PGD | accuracy: -3.8±0.3 | Helped |
| add high SF Gabors (ablation) | VOneResNet50 · ImageNet-C | accuracy: 1.9±0.4 | Hurt |
| add complex cells (ablation) | VOneResNet50 · ImageNet white box PGD | accuracy: -4.5±1.1 | Helped |
| add simple cells (ablation) | VOneResNet50 · ImageNet-C | accuracy: -4.8±0.5 | Helped |
| use primate V1 Gabor parameter distributions instead of uniform (ablation) | VOneResNet50 · ImageNet perturbation benchmark | overall mean accuracy: -0.4±0.3 | No clear effect |
| add V1 stochasticity alone to ResNet50 first block (ablation) | ResNet50 · ImageNet perturbation benchmark | perturbation mean accuracy | Helped |
Closest papers by meaning
- Explicitly Modeling Subcortical Vision with a Neuro-Inspired Front-End Improves CNN Robustness (NeurIPS 2025 · similarity 0.93)
- Towards robust vision by multi-task learning on monkey visual cortex (NeurIPS 2021 · similarity 0.91)
- Performance-optimized deep neural networks are evolving into worse models of inferotemporal visual cortex (NeurIPS 2023 · similarity 0.90)
- Explaining V1 Properties with a Biologically Constrained Deep Learning Architecture (NeurIPS 2023 · similarity 0.89)
- Brain-Like Object Recognition with High-Performing Shallow Recurrent ANNs (NeurIPS 2019 · similarity 0.88)
- Benchmarking Out-of-Distribution Generalization Capabilities of DNN-based Encoding Models for the Ventral Visual Cortex. (NeurIPS 2024 · 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.