Research map / Learning theory and algorithms / Matrix and tensor methods
Trading Positional Complexity vs Deepness in Coordinate Networks
Jianqiao Zheng, Sameera Ramasinghe, Xueqian Li, Simon Lucey
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 simple encoding + MLP with complex Kronecker encoding + single linear layer (Gau) | Gaussian embedder · 2D image, separable | PSNR (dB): 26.17 -> 26.69 | Improved |
| replace simple encoding + MLP with complex encoding (Tri) | Triangle embedder · 2D image, separable | PSNR (dB): 25.96 -> 26.65 | Improved |
| use complex encoding instead of simple encoding for frequency features | LinF · 2D image, separable | PSNR (dB): 26.12 -> 19.73 | Hurt |
| use complex encoding with a single linear layer | RFF · 2D image, separable | PSNR (dB): 26.58 -> 24.34 | Hurt |
| solve complex encoding in closed form instead of gradient descent | Gaussian embedder · 2D image, separable | training time (s): 0.18 | Improved |
| use complex encoding instead of simple encoding (training time) | Gaussian embedder · 2D image, separable | training time (s): 23.12 -> 2.01 | Improved |
| use complex encoding with a single linear layer | Gaussian embedder · 2D image, random 25% | PSNR (dB): 24.63 -> 23.97 | Hurt |
| use complex encoding with a single linear layer | Triangle embedder · 2D image, random 25% | PSNR (dB): 24.24 -> 23.21 | Hurt |
| use complex encoding with a single linear layer | Triangle embedder · video | PSNR (dB): 20.86 -> 22.24 | Improved |
| use complex encoding with a single linear layer | LinF · video | PSNR (dB): 21.49 -> 9.76 | Hurt |
| solve complex encoding in closed form | Triangle embedder · video | training time (s): 78.91 -> 0.009 | Improved |
| add one hidden layer to a linear network when embedding rank is low (ablation) | Gaussian embedder · 1D signal | reconstruction performance | Improved |
Closest papers by meaning
- A Learnable Radial Basis Positional Embedding for Coordinate-MLPs (AAAI 2023 · similarity 0.91)
- Learnable Spatial-Temporal Positional Encoding for Link Prediction (ICML 2025 · similarity 0.85)
- RelFlexformer: Efficient Attention Transformers for Integrable Relative Positional Encodings (NeurIPS 2026 · similarity 0.85)
- On the Stability of Expressive Positional Encodings for Graphs (ICLR 2024 · similarity 0.85)
- Positional Attention: Expressivity and Learnability of Algorithmic Computation (ICML 2025 · similarity 0.84)
- ComRoPE: Scalable and Robust Rotary Position Embedding Parameterized by Trainable Commuting Angle Matrices (CVPR 2025 · similarity 0.84)
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.
