Research map / Architectures and efficiency / CNN design
Powerset Convolutional 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 |
|---|---|---|---|
| use powerset convolutional layers instead of the strongest reported baseline | set-function classification models · Patterns | test classification accuracy (%): 65.5 ± 0.9 -> 88.5 ± 4.3 | Helped |
| use powerset convolutional layers instead of the strongest reported baseline | set-function classification models · k-Junta | test classification accuracy (%): 95.8 ± 2.7 -> 97.5 ± 1.4 | Helped |
| use powerset convolutional layers instead of the strongest reported baseline | set-function classification models · Submodularity | test classification accuracy (%): 89.8 ± 1.8 -> 88.6 ± 0.4 | Hurt |
| use powerset convolutional layers instead of the strongest reported baseline | hypergraph classification models · COAUTH10 | test classification accuracy (%): 84.7 ± 0.9 -> 84.8 ± 0.3 | No clear effect |
| use powerset convolutional layers instead of the strongest reported baseline | hypergraph classification models · CON10 | test classification accuracy (%): 67.2 ± 1.8 -> 68.7 ± 1.3 | Helped |
| use powerset convolutional layers instead of the strongest reported baseline | hypergraph classification models · DOM4 | test classification accuracy (%): 96.0 ± 0.2 -> 94.8 ± 0.3 | Hurt |
| use powerset convolutional layers instead of the strongest reported baseline | hypergraph classification models · DOM6 | test classification accuracy (%): 73.7 ± 0.4 -> 73.5 ± 0.5 | No clear effect |
| add pooling after each convolutional layer (ablation) | *-PCN · Patterns | test classification accuracy (%): 88.5 ± 4.3 -> 80.9 ± 0.9 | Hurt |
| add pooling and feature accumulation before the MLP (ablation) | *-PCN · Patterns | test classification accuracy (%): 88.5 ± 4.3 -> 75.9 ± 1.9 | Hurt |
| add pooling and feature accumulation before the MLP (ablation) | diamond-PCN · k-Junta | test classification accuracy (%): 97.5 ± 1.4 -> 96.6 ± 0.7 | Hurt |
| add feature accumulation before the MLP to make the convolutional part shift-invariant (ablation) | powerset CNNs · Submodularity | test classification accuracy (%): 88.6 ± 0.4 -> 87.0 ± 1.6; 80.9 ± 2.9 | Mixed |
| add pooling and feature accumulation before the MLP (ablation) | *-PCN · DOM4 | test classification accuracy (%): 94.1 ± 0.3 -> 94.4 ± 0.3 | No clear effect |
Closest papers by meaning
- Joint Learning of Set Cardinality and State Distribution (AAAI 2018 · similarity 0.86)
- Polynomial Width is Sufficient for Set Representation with High-dimensional Features (ICLR 2024 · similarity 0.85)
- Multiset-Equivariant Set Prediction with Approximate Implicit Differentiation (ICLR 2022 · similarity 0.85)
- Generalizing CNNs to graphs with learnable neighborhood quantization (NeurIPS 2024 · similarity 0.83)
- Equivariant Machine Learning on Graphs with Nonlinear Spectral Filters (NeurIPS 2024 · similarity 0.83)
- Graph As Point Set (ICML 2024 · 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.