Research map / Adversarial robustness and security / Certified and empirical robustness
Geometric Robustness of Deep Networks: Analysis and Improvement
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 ManiFool instead of Manitest to find fooling transformations | 2-layer CNN · MNIST, translation (T) | time per sample: 2.7 s -> 2.6 s | Helped |
| Use ManiFool instead of Manitest to find fooling transformations | 2-layer CNN · MNIST, rotation and translation (R+T) | time per sample: 23.9 s -> 3.6 s | Helped |
| Use ManiFool instead of Manitest to find fooling transformations | 2-layer CNN · MNIST, scaling and translation (S+T) | time per sample: 34.6 s -> 6.2 s | Helped |
| Use ManiFool instead of Manitest to find fooling transformations | 2-layer CNN · MNIST, translation, rotation and scaling (T+R+S) | time per sample: 29.5 s -> 3.1 s | Helped |
| Fine-tune using ManiFool-generated minimal label-changing affine transformations | ResNet-18 · CIFAR10 test | invariance score: 1.13 -> 1.78 | Helped |
| Fine-tune using randomly transformed images | ResNet-18 · CIFAR10 test | invariance score: 1.13 -> 1.55 | Helped |
| Fine-tune using ManiFool-generated worst-case transformations rather than random transformations | ResNet-18 · CIFAR10 test | invariance score: 1.55 -> 1.78 | Helped |
| Fine-tune using ManiFool-generated worst-case transformations | ResNet-18 · CIFAR10 test, random affine transformations at distance 2.5 | misclassification rate: decreased more than 20% | Helped |
| Fine-tune using ManiFool-generated worst-case transformations | ResNet-18 · CIFAR10 test | accuracy: 0.6% reduction | Hurt |
Closest papers by meaning
- DeepFool: A Simple and Accurate Method to Fool Deep Neural Networks (CVPR 2016 · similarity 0.87)
- Certifying Geometric Robustness of Neural Networks (NeurIPS 2019 · similarity 0.86)
- Geometric Inductive Biases of Deep Networks: The Role of Data and Architecture (ICLR 2025 · similarity 0.85)
- Towards Verifying the Geometric Robustness of Large-Scale Neural Networks (AAAI 2023 · similarity 0.85)
- Asymptotics of Learning with Deep Structured (Random) Features (ICML 2024 · similarity 0.85)
- On the Completeness of Invariant Geometric Deep Learning Models (ICLR 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.