Research map / Adversarial robustness and security / Attacks on text and graphs
Degraded Polygons Raise Fundamental Questions of Neural Network Perception
Leonard Tang, Dan Ley
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (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 ImageNet rather than FractalDB pretraining | ResNet-18 · Pentagon, 0.5 Corner Degradation | per-class accuracy (%): 100.0 -> 0.1 | Hurt |
| Use FractalDB rather than ImageNet pretraining | ResNet-18 · Pentagon, 0.5 Edge Degradation | per-class accuracy (%): 0.0 -> 2.4 | Improved |
| Degrade polygon edges rather than corners | ImageNet-pretrained ResNet-18 · Pentagon, 0.5 degradation | per-class accuracy (%): 0.1 -> 0.0 | No clear effect |
| Degrade polygon edges rather than corners | FractalDB-pretrained ResNet-18 · Pentagon, 0.5 degradation | per-class accuracy (%): 100.0 -> 2.4 | Hurt |
| Increase perimeter degradation | Vision architectures · Degraded regular polygons | top-1 test accuracy | Hurt |
| Use FractalDB rather than ImageNet pretraining | ResNet-18 and ResNet-50 · Degraded regular polygons | top-1 test accuracy | Mixed |
| Use ResNet-18, ResNet-50, or MLP-Mixer instead of ViT | ImageNet-pretrained models · Edge- versus corner-degraded polygons | top-1 test accuracy | Mixed |
Closest papers by meaning
- Generalisation in humans and deep neural networks (NeurIPS 2018 · similarity 0.88)
- ImageNet-trained CNNs are biased towards texture; increasing shape bias improves accuracy and robustness (ICLR 2019 · similarity 0.86)
- Progress and Limitations of Deep Networks to Recognize Objects in Unusual Poses (AAAI 2023 · similarity 0.86)
- Recognizing Object by Components With Human Prior Knowledge Enhances Adversarial Robustness of Deep Neural Networks (TPAMI 2023 · similarity 0.86)
- Towards robust vision by multi-task learning on monkey visual cortex (NeurIPS 2021 · similarity 0.86)
- Contour Integration Underlies Human-Like Vision (ICML 2025 · similarity 0.85)
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.
