Research map / Probabilistic and scientific modeling / Molecules and proteins
NCDL: A Framework for Deep Learning on non-Cartesian Lattices
Joshua John Horacsek, Usman Alim
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 |
|---|---|---|---|
| Replace Cartesian convolutions with quincunx downsampling plus quincunx convolution | convolutional auto-encoder · CelebA | PSNR: 17.53 -> 17.92 | Improved |
| Replace Cartesian convolutions with quincunx downsampling plus quincunx convolution | convolutional auto-encoder · CelebA | L1: 0.08804 -> 0.08318 | Improved |
| Replace Cartesian convolutions with quincunx downsampling plus quincunx convolution | convolutional auto-encoder · CelebA | SSIM: 0.6113 -> 0.6206 | Improved |
| Replace Cartesian convolutions with quincunx downsampling plus quincunx convolution | convolutional auto-encoder · CelebA | LPIPS: 0.4808 -> 0.4765 | Improved |
| Replace Cartesian with quincunx lattice in U-Net | U-Net · DUTS | BCE: 0.19266 -> 0.18706 | Improved |
| Replace Cartesian with quincunx lattice in U-Net | U-Net · DUTS | SSIM: 0.2900 -> 0.3074 | Improved |
| Replace Cartesian with quincunx lattice in U-Net | U-Net · DUTS | L1: 0.07583 -> 0.08268 | Hurt |
| Replace Cartesian with quincunx lattice in U-Net | U-Net · DUTS | L2: 0.04969 -> 0.05107 | Hurt |
| Use NCDL hexagonal convolution instead of HexagDLy at small grid sizes | hexagonal convolution · GPU runtime, grid 256-2048 | mean time (ms) | Hurt |
| Use NCDL hexagonal convolution instead of HexagDLy at large grid sizes | hexagonal convolution · GPU runtime, grid 256-2048 | mean time (ms) | Improved |
| Use NCDL non-Cartesian convolution instead of Cartesian Conv2D | 1-ring filter · GPU runtime | mean time (ms) | Mixed |
Closest papers by meaning
- Low-Rank Tensor Function Representation for Multi-Dimensional Data Recovery (TPAMI 2023 · similarity 0.84)
- Convolutions and More as Einsum: A Tensor Network Perspective with Advances for Second-Order Methods (NeurIPS 2024 · similarity 0.84)
- Non-Convex Projected Gradient Descent for Generalized Low-Rank Tensor Regression (JMLR 2019 · similarity 0.84)
- Nonparametric Decomposition of Sparse Tensors (ICML 2021 · similarity 0.83)
- Tensor learning with orthogonal, Lorentz, and symplectic symmetries (ICLR 2026 · similarity 0.83)
- Efficient Nonparametric Tensor Decomposition for Binary and Count Data (AAAI 2024 · similarity 0.82)
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.
