Research map / Architectures and efficiency / CNN design
CKConv: Continuous Kernel Convolution For Sequential Data
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 continuous convolution kernels parameterized by Sine networks instead of discrete kernels | CKCNN · Copy Memory, sequences up to length 6000 | solved task | Helped |
| Use continuous convolution kernels parameterized by Sine networks instead of discrete kernels | CKCNN · pMNIST test | accuracy: 98.54 vs 98.30 | Helped |
| Use continuous convolution kernels parameterized by Sine networks instead of discrete kernels | CKCNN · sCIFAR10 test | accuracy: 63.74 vs 73.42 | Hurt |
| Use continuous convolution kernels parameterized by Sine networks instead of discrete kernels | CKCNN · Char-PTB test | bpc: 1.158 -> 1.045 | Helped |
| Use continuous convolution kernels parameterized by Sine networks instead of discrete kernels | CKCNN · PhysioNet test | AUC: 0.880 -> 0.895 | Helped |
| Use continuous convolution kernels parameterized by Sine networks instead of discrete kernels | CKCNN · CharacterTrajectories, 70% data dropped | accuracy: 98.6 vs 98.14 | Hurt |
| Use Sine nonlinearities in the kernel network rather than ReLU, LeakyReLU, or Swish (ablation) | CKCNN kernel parameterization · Function approximation and considered tasks | approximation capability and task performance | Helped |
| Increase CKCNN depth beyond two residual blocks (ablation) | CKCNN · Experiments reported in the paper | performance | No clear effect |
| Use a predefined shared omega_0 rather than learning layer-wise omega_0 values (ablation) | CKCNN · Experiments reported in the paper | performance | Helped |
| Tune omega_0 across values in [1, 100] (ablation) | CKCNN on pMNIST · pMNIST | accuracy: 98.54 to 65.22 | Mixed |
| Use uniformly distributed Sine-network bias initialization (ablation) | CKCNN kernel network · All tasks considered | performance and convergence | Helped |
| Use CKCNN parallel convolution instead of NCDE on SC_raw | Sequential modeling · SpeechCommands_raw | training speed: 228× | Helped |
Closest papers by meaning
- FlexConv: Continuous Kernel Convolutions With Differentiable Kernel Sizes (ICLR 2022 · similarity 0.88)
- A Temporal Kernel Approach for Deep Learning with Continuous-time Information (ICLR 2021 · similarity 0.88)
- KernelDNA: Cross-Layer Kernel Sharing via Decoupled Neural Adapters (NeurIPS 2026 · similarity 0.87)
- Reparameterized Multi-Resolution Convolutions for Long Sequence Modelling (NeurIPS 2024 · similarity 0.86)
- KernelWarehouse: Rethinking the Design of Dynamic Convolution (ICML 2024 · similarity 0.85)
- Continuity-Preserving Convolutional Autoencoders for Learning Continuous Latent Dynamical Models from Images (ICLR 2025 · 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.