Research map / Detection and segmentation / CNN features
AOGNets: Compositional Grammatical Architectures for Deep Learning
Xilai Li, Xi Song, Tianfu Wu
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 an AOG building-block structure instead of conventional blocks, with the same bottleneck node operation | ImageNet-1K classification · ImageNet-1K val | top-1 error (%): 23.0 -> 19.34 | Improved |
| Use an AOGNet backbone instead of ResNet-101 | Mask R-CNN C4 · coco_val2017 | box AP: 39.2 -> 41.4 | Improved |
| Use an AOGNet backbone instead of ResNeXt-101 | Mask R-CNN FPN · coco_val2017 | box AP: 42.2 -> 42.5 | Improved |
| Use AOGNet-60M rather than ResNet-152 | ImageNet-1K white-box FGSM · ε=0.1 | top-1 accuracy: 16.3 -> 30.2 | Improved |
| Use AOGNet-60M rather than ResNet-152 | ImageNet-1K white-box FGSM · ε=0.3 | top-1 accuracy: 0.85 -> 2.6 | Improved |
| Use AOGNet-4M rather than MobileNetV2 (1.4) | ImageNet-1K mobile settings · ImageNet-1K val | top-1 error (%): 25.3 -> 26.2 | Hurt |
| Add lateral connections (ablation) | AOGNet · CIFAR-10 | error rate (%): 3.75 -> 3.70 | Improved |
| Add lateral connections (ablation) | AOGNet · CIFAR-100 | error rate (%): 19.20 -> 19.09 | Improved |
| Remove symmetric OR-node children (ablation) | AOGNet · CIFAR-10 | error rate (%): 3.75 -> 3.57 | Improved |
| Remove symmetric OR-node children (ablation) | AOGNet · CIFAR-100 | error rate (%): 19.20 -> 18.64 | Improved |
| Combine symmetric-child removal and lateral connections (ablation) | AOGNet · CIFAR-10 | error rate (%): 3.75 -> 3.52 | Improved |
| Combine symmetric-child removal and lateral connections (ablation) | AOGNet · CIFAR-100 | error rate (%): 19.20 -> 17.99 | Improved |
Closest papers by meaning
- Blended Grammar Network for Human Parsing (ECCV 2020 · similarity 0.85)
- Learning to Recombine and Resample Data For Compositional Generalization (ICLR 2021 · similarity 0.85)
- Deep networks learn to parse uniform-depth context-free languages from local statistics (ICML 2026 · similarity 0.85)
- Consistency of Compositional Generalization Across Multiple Levels (AAAI 2025 · similarity 0.83)
- Group and Shuffle: Efficient Structured Orthogonal Parametrization (NeurIPS 2024 · similarity 0.83)
- Causal Graphical Models for Vision-Language Compositional Understanding (ICLR 2025 · similarity 0.83)
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.
