Research map / Architectures and efficiency / CNN design
On the Expressive Power of Overlapping Architectures of Deep Learning
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 overlapping receptive fields instead of non-overlapping fields | ConvACs · Expressive-capacity analysis | grid-tensor matricization rank: D^{\left\lfloor \frac{H - T_R^{(K, \left\lfloor \nicefrac{H}{2} \right\rfloor)}}{T_S^{(K)}} +1 \right\rfloor \cdot \left | Helped |
| Add limited overlap using B×B stride-1 layers followed by 2×2 stride-2 layers | ConvACs · Theoretical lower bound for expressive capacity | non-overlapping network size required to match: M^{\frac{(2B-1)^2}{2} \cdot \left(1+\frac{2B-2}{H}\right)^{-2}} | Helped |
| Use B=5 overlapping layers with M=64 and H ≥ 20 | ConvACs · Theoretical lower bound for expressive capacity | non-overlapping network size required to match: 64^{20} | Helped |
| Use overlapping rather than non-overlapping convolutions | Standard ConvNets · CIFAR-10 spatially augmented training set | channels required for similar training accuracy: more than 2048 channels -> 64 channels | Helped |
| Use overlapping convolutions rather than non-overlapping convolutions | Standard ConvNets · CIFAR-10 color-augmented training set | training accuracy | Helped |
| Increase receptive-field size B | Standard ConvNets · CIFAR-10 augmented training sets | channels required for a given training accuracy | Helped |
| Increase overlap beyond the smallest overlapping case | Standard ConvNets · CIFAR-10 augmented training sets | training accuracy at the same parameter count | No clear effect |
Closest papers by meaning
- Boosting Dilated Convolutional Networks with Mixed Tensor Decompositions (ICLR 2018 · similarity 0.94)
- The Expressive Power of Neural Networks: A View from the Width (NeurIPS 2017 · similarity 0.89)
- Exploring the Complexity of Deep Neural Networks through Functional Equivalence (ICML 2024 · similarity 0.85)
- On the Expressiveness of Rational ReLU Neural Networks With Bounded Depth (ICLR 2025 · similarity 0.84)
- The impact of allocation strategies in subset learning on the expressive power of neural networks (ICLR 2025 · similarity 0.84)
- Scaling Laws and Tradeoffs in Recurrent Networks of Expressive Neurons (NeurIPS 2026 · similarity 0.84)
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.