Research map / Learning theory and algorithms / Kernels and regression theory
Approximation and Learning with Deep Convolutional Models: a Kernel Perspective
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 a second-layer Poly3 kernel instead of Exp (ablation) | 2-layer CKN · CIFAR-10 test (50k training) | accuracy: 88.3% -> 88.2% | No clear effect |
| Use a second-layer Poly2 kernel instead of Exp (ablation) | 2-layer CKN · CIFAR-10 test (50k training) | accuracy: 88.3% -> 87.4% | Hurt |
| Use a second-layer Poly2 kernel instead of Exp (ablation) | 2-layer CKN · SVHN test | accuracy: 89.5% -> 88.6% | Hurt |
| Use a linear kernel at the second layer instead of a nonlinear kernel (ablation) | 2-layer CKN · SVHN test | accuracy: 89.5% -> 78.5% | Hurt |
| Use Poly2 in the first layer instead of Exp (ablation) | 2-layer CKN · CIFAR-10 test (10k training) | accuracy: 87.9% -> 85.1% | Hurt |
| Remove the second nonlinear convolutional layer (ablation) | 2-layer CKN · CIFAR-10 test (10k training) | accuracy: 87.9% -> 80.9% | Hurt |
| Use degree-2 kernels at both higher layers instead of Exp (ablation) | 3-layer CKN · CIFAR-10 test (50k training) | accuracy: 88.2% -> 87.9% | Hurt |
| Remove Gaussian pooling by using strided pooling (ablation) | CKN · CIFAR-10 test (10k training) | accuracy: less than 40% | Hurt |
| Replace Gaussian pooling with average pooling (ablation) | 2-layer CKN (Exp, Exp), CIFAR-10 · CIFAR-10 test (50k training) | accuracy: 88.3% -> 75.9% | Hurt |
| Replace Gaussian pooling with average pooling (ablation) | 3-layer CKN (Exp, Exp, Exp) · CIFAR-10 test (50k training) | accuracy: 88.2% -> 72.4% | Hurt |
| Increase first-layer patch size from 3x3 to 5x5 (ablation) | 2-layer CKN · CIFAR-10 test (10k training) | accuracy: 80.5% -> 79.6% | Hurt |
| Use global ZCA instead of patch-level whitening (ablation) | 2-layer CKN · CIFAR-10 test (50k training) | accuracy: 88.3% -> about 80% | Hurt |
Closest papers by meaning
- KernelWarehouse: Rethinking the Design of Dynamic Convolution (ICML 2024 · similarity 0.91)
- Learning with convolution and pooling operations in kernel methods (NeurIPS 2022 · similarity 0.91)
- CNN Filter DB: An Empirical Investigation of Trained Convolutional Filters (CVPR 2022 · similarity 0.91)
- The Unreasonable Effectiveness of Patches in Deep Convolutional Kernels Methods (ICLR 2021 · similarity 0.90)
- Unveiling the Unseen: Identifiable Clusters in Trained Depthwise Convolutional Kernels (ICLR 2024 · similarity 0.89)
- UniRepLKNet: A Universal Perception Large-Kernel ConvNet for Audio Video Point Cloud Time-Series and Image Recognition (CVPR 2024 · similarity 0.88)
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.