Research map / Architectures and efficiency / CNN design
Learning from Few Samples: Transformation-Invariant SVMs with Composition and Locality at Multiple Scales
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 |
|---|---|---|---|
| Add locality at multiple spatial scales and translation/rotation invariance | SVM · Original MNIST, 100/200/500 training samples | accuracy (%): 68.33%/83.20%/91.33% -> 81.55%/89.23%/93.11% | Helped |
| Add locality at multiple spatial scales and translation/rotation invariance | SVM · EMNIST Letters, 100/200/500 training samples | accuracy (%): 36.65/42.74/56.38 -> 44.56/55.18/66.42 | Helped |
| Add locality at multiple spatial scales (ablation) | TI-RI-SVM · Original MNIST, 100/200 training samples | accuracy (%): 75.10 -> 81.55; 86.47 -> 89.23 | Helped |
| Add locality at multiple spatial scales (ablation) | TI-RI-SVM · Original MNIST, 500 training samples | accuracy (%): 93.11 -> 92.58 | Hurt |
| Add locality at multiple spatial scales (ablation) | TI-RI-SVM · EMNIST Letters, 100/200 training samples | accuracy (%): 43.16 -> 44.56; 52.40 -> 55.18 | Helped |
| Add locality at multiple spatial scales (ablation) | TI-RI-SVM · EMNIST Letters, 500 training samples | accuracy (%): 67.42 -> 66.42 | Hurt |
| Add rotation invariance (ablation) | L-TI-SVM · Original MNIST, 100/200 training samples | accuracy (%): 78.86 -> 81.55; 87.02 -> 89.23 | Helped |
| Add rotation invariance (ablation) | L-TI-SVM · Original MNIST, 500 training samples | accuracy (%): 91.01 -> 92.58 | Helped |
| Add rotation invariance (ablation) | L-TI-SVM · EMNIST Letters, 100/200/500 training samples | accuracy (%): 42.51 -> 44.56; 52.81 -> 55.18; 64.66 -> 66.42 | Helped |
| Use best-fit rather than average-fit transformation similarity | translation-invariant SVM · Original MNIST and EMNIST Letters | accuracy | Helped |
| Use best-fit rather than average-fit transformation similarity | translation-invariant SVM · Original MNIST, 100/200/500 training samples | accuracy (%): TI-SVM 69.34/82.34/91.00; RI-SVM (Average-Fit) 68.05/78.81/87.21 | Mixed |
| Use transformation-invariant kernels | SVM · Original MNIST and EMNIST Letters | positive definiteness: Original MNIST: approximately 45 -> approximately 375; EMNIST: approximately 35 -> approximately 395 | Helped |
Closest papers by meaning
- Exploiting Invariance in Training Deep Neural Networks (AAAI 2022 · similarity 0.88)
- Learning Invariances in Neural Networks from Training Data (NeurIPS 2020 · similarity 0.88)
- Max-Margin Invariant Features from Transformed Unlabelled Data (NeurIPS 2017 · similarity 0.87)
- Role of Locality and Weight Sharing in Image-Based Tasks: A Sample Complexity Separation between CNNs, LCNs, and FCNs (ICLR 2024 · similarity 0.87)
- On Making $SE(2)$-Invariant Networks Optimal (NeurIPS 2026 · similarity 0.85)
- From Kernels to Features: A Multi-Scale Adaptive Theory of Feature Learning (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.