Research map / Architectures and efficiency / CNN design
Perturbative Neural Networks
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 standard convolutional layers with perturbation layers | ResNet-18 · ImageNet-1k validation | top-1 accuracy: 69.57 -> 71.84 | Helped |
| Replace standard convolutional layers with perturbation layers | ResNet-50 · ImageNet-1k validation | top-1 accuracy: 75.99 -> 76.23 | Helped |
| Replace convolution layers with perturbation layers | Faster R-CNN with VGG-16 · PASCAL VOC 2007 test | mAP: 70.2 -> 69.6 | Hurt |
| Replace convolution layers with perturbation layers | Faster R-CNN with ResNet-50 · PASCAL VOC 2007 test | mAP: 75.2 -> 72.8 | Hurt |
| Increase perturbation masks from 64 to 256 per layer (ablation) | PNN-ResNet-18 · ImageNet-1k validation | top-1 accuracy: 45.92 -> 71.84 | Helped |
| Increase perturbation noise range from 0.5 to 1 (ablation) | PNN-ResNet-50 · ImageNet-100 | top-1 accuracy: 81.96 -> 76.84 | Hurt |
| Increase perturbation noise range from 1 to 5 (ablation) | PNN-ResNet-50 · ImageNet-100 | top-1 accuracy: 76.84 -> 60.90 | Hurt |
| Vary perturbation mask count from 8 to 160 per layer (ablation) | PNN-ResNet-18 · CIFAR-10 | accuracy: 79.98 to 93.08 across mask counts | Mixed |
| Reduce batch size from 80 to 10 with 64 masks per layer (ablation) | PNN-ResNet-50 · CIFAR-10 | accuracy: 85.29 -> 94.05 | Helped |
Closest papers by meaning
- Deep Perturbation Learning: Enhancing the Network Performance via Image Perturbations (ICML 2023 · similarity 0.89)
- Learning perturbation sets for robust machine learning (ICLR 2021 · similarity 0.88)
- Verification of Neural Networks Against Convolutional Perturbations via Parameterised Kernels (AAAI 2025 · similarity 0.87)
- Contrasting Adversarial Perturbations: The Space of Harmless Perturbations (AAAI 2025 · similarity 0.86)
- TrAct: Making First-layer Pre-Activations Trainable (NeurIPS 2024 · similarity 0.86)
- Wide Two-Layer Networks can Learn from Adversarial Perturbations (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.