Research map / Architectures and efficiency / CNN design
Gaussian Mixture Convolution Networks
Adam Celarek, Pedro Hermosilla, Bernhard Kerbl, Timo Ropinski, Michael Wimmer
Read the paperCheck your idea against 137,217 papers
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 |
|---|---|---|---|
| Represent inputs and convolution kernels as Gaussian mixtures and use analytical convolution | GMCN · MNIST test | accuracy: 99.2\,\% -> 99.4\,\% | Improved |
| Represent 3D inputs as Gaussian mixtures and use GMCN | 3D classification · ModelNet10 test | accuracy: 92.2\% -> 93.3\% | Mixed |
| Increase depth from 1 to 5 convolution layers (ablation) | GMCN · ModelNet10 | accuracy: 72.5\,\% -> 92.6\,\% | Improved |
| Increase depth from 5 to 6 convolution layers (ablation) | GMCN · ModelNet10 | accuracy: 92.6\,\% -> 92.3\,\% | Hurt |
| Use TreeHEM (T=2) instead of modified EM for reduction fitting (ablation) | GMCN · ModelNet10, 64 input Gaussians | accuracy: 92.2\,\% -> 92.0\,\% | Hurt |
| Use TreeHEM (T=2) instead of TreeHEM (T=4) (ablation) | GMCN · ModelNet10, 128 input Gaussians | accuracy: 92.9\,\% -> 93.4\,\% | Improved |
| Use TreeHEM reduction rather than modified EM (ablation) | GMCN · Convolution fitting, layers 1–3 | RMSE: 0.45 -> 0.36; 0.50 -> 0.21; 0.36 -> 0.22 | Improved |
| Fit input GMs with k-means initialization and one EM step rather than k-means+EM (ablation) | GMCN · ModelNet10 | accuracy: 92.1\,\% -> 93.3\,\% | Improved |
| Fit input GMs with random initialization and multiple EM steps rather than k-means (ablation) | GMCN · ModelNet10 | accuracy: 93.3\,\% -> 92.8\,\% | Hurt |
| Fit input GMs with Eckart rather than k-means (ablation) | GMCN · ModelNet10 | accuracy: 93.3\,\% -> 91.0\,\% | Hurt |
| Reduce input representation from 128 to 32 Gaussians without changing internal fitting (ablation) | GMCN · ModelNet10 | accuracy: 92.78 -> 91.44 | Hurt |
Closest papers by meaning
- Approximation and Learning with Deep Convolutional Models: a Kernel Perspective (ICLR 2022 · similarity 0.90)
- Classifying high-dimensional Gaussian mixtures: Where kernel methods fail and neural networks succeed (ICML 2021 · similarity 0.88)
- Deep Convolutional Networks as shallow Gaussian Processes (ICLR 2019 · similarity 0.87)
- Feature learning from non-Gaussian inputs: the case of Independent Component Analysis in high dimensions (ICML 2025 · similarity 0.86)
- Asymptotic Analysis of Two-Layer Neural Networks after One Gradient Step under Gaussian Mixtures Data with Structure (ICLR 2025 · similarity 0.85)
- Unveiling the Unseen: Identifiable Clusters in Trained Depthwise Convolutional Kernels (ICLR 2024 · 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.
