Research map / Architectures and efficiency / CNN design
The Unreasonable Effectiveness of Patches in Deep Convolutional Kernels Methods
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 patches sampled from training data instead of random Gaussian patches (ablation) | SimplePatch · CIFAR-10 | accuracy: 8.1% | Helped |
| replace hard-assignment VQ with soft-assignment VQ (ablation) | SimplePatch · CIFAR-10 | accuracy: 0.2% | Helped |
| train the patch dictionary end-to-end rather than sampling patches (ablation) | SimplePatch · CIFAR-10 | accuracy: - 0.1 % | No clear effect |
| increase image resolution from 64 to 128 (ablation) | SimplePatch · ImageNet | Top-1 accuracy: 33.2 -> 35.9 | Helped |
| omit hard-assignment VQ (ablation) | SimplePatch at resolution 128 · ImageNet | Top-1 accuracy: 35.9 -> 36.0 | No clear effect |
| replace CIFAR-10-sampled patches with ImageNet-sampled patches (ablation) | SimplePatch · CIFAR-10 | accuracy: 0.4% | Hurt |
| remove data augmentation (ablation) | SimplePatch · CIFAR-10 | accuracy: about 2% | Hurt |
| vary dictionary size, nearest-neighbor count, patch size, and whitening regularization (ablation) | SimplePatch · CIFAR-10 | accuracy | No clear effect |
Closest papers by meaning
- End-to-End Kernel Learning with Supervised Convolutional Kernel Networks (NeurIPS 2016 · similarity 0.91)
- Approximation and Learning with Deep Convolutional Models: a Kernel Perspective (ICLR 2022 · similarity 0.90)
- Stochastic Kernel Regularisation Improves Generalisation in Deep Kernel Machines (NeurIPS 2024 · similarity 0.88)
- KernelWarehouse: Rethinking the Design of Dynamic Convolution (ICML 2024 · similarity 0.87)
- SELECT: A Large-Scale Benchmark of Data Curation Strategies for Image Classification (NeurIPS 2024 · similarity 0.86)
- Convolutional Deep Kernel Machines (ICLR 2024 · similarity 0.86)
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.