Research map / Architectures and efficiency / CNN design
Convexified 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 CCNN-2 instead of the strongest listed baseline | MNIST basic · MNIST basic | classification error: 1.06% -> 1.38% | Hurt |
| Use CCNN-2 instead of CNN-2 (ReLU) | MNIST rand · MNIST rand | classification error: 5.64% -> 4.32% | Helped |
| Use CCNN-2 instead of TIRBM | MNIST rot · MNIST rot | classification error: 4.20% -> 6.98% | Hurt |
| Use CCNN-2 instead of CNN-2 (ReLU) | MNIST img · MNIST img | classification error: 10.17% -> 7.46% | Helped |
| Use CCNN-2 instead of PCANet-2 | MNIST img+rot · MNIST img+rot | classification error: 35.48% -> 30.23% | Helped |
| Convexify the top convolutional layer using CCNN | CNN models · CIFAR-10 | classification error: 34.14% -> 23.62%; 24.98% -> 21.88%; 21.48% -> 18.18% | Helped |
| Replace the Gaussian kernel with a linear kernel (ablation) | CCNN-2 · MNIST img+rot | classification error: 30.23% -> 38.47% | Hurt |
| Remove ZCA whitening (ablation) | CCNN-2 · MNIST img+rot | classification error: 30.23% -> 35.86% | Hurt |
| Use fewer random features (m=200 rather than m=500) (ablation) | CCNN-2 · MNIST img+rot | classification error: 30.23% -> 31.95% | Hurt |
| Replace the nuclear norm constraint with a Frobenius norm constraint (ablation) | CCNN-2 · MNIST img+rot | classification error: 30.23% -> 30.66% | No clear effect |
| Stop mini-batch SGD before convergence (ablation) | CCNN-2 · MNIST img+rot | classification error: 30.23% -> 30.63% | No clear effect |
Closest papers by meaning
- Implicit Convex Regularizers of CNN Architectures: Convex Optimization of Two- and Three-Layer Networks in Polynomial Time (ICLR 2021 · similarity 0.89)
- Deeply Shared Filter Bases for Parameter-Efficient Convolutional Neural Networks (NeurIPS 2021 · similarity 0.87)
- Neural Networks are Convex Regularizers: Exact Polynomial-time Convex Optimization Formulations for Two-layer Networks (ICML 2020 · similarity 0.87)
- Fast Convex Optimization for Two-Layer ReLU Networks: Equivalent Model Classes and Cone Decompositions (ICML 2022 · similarity 0.87)
- The Hidden Convex Optimization Landscape of Regularized Two-Layer ReLU Networks: an Exact Characterization of Optimal Solutions (ICLR 2022 · similarity 0.87)
- DCFNet: Deep Neural Network with Decomposed Convolutional Filters (ICML 2018 · similarity 0.87)
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.