Research map / Architectures and efficiency / Attention and transformers
Attention-based Neural Cellular Automata
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 ViTCA instead of the strongest listed CA baseline (UNetCA) | Denoising autoencoding · MNIST | PSNR (dB): 25.61 -> 27.01 | Helped |
| Use ViTCA instead of the strongest listed CA baseline (UNetCA) | Denoising autoencoding · CelebA | PSNR (dB): 25.66 -> 26.53 | Helped |
| Use ViTCA instead of the strongest listed CA baseline (UNetCA) | Denoising autoencoding · LandCoverRep | PSNR (dB): 33.94 -> 33.80 | Hurt |
| Use ViTCA instead of the strongest listed CA baseline (UNetCA) | Denoising autoencoding · FashionMNIST | PSNR (dB): 23.67 -> 23.80 | Helped |
| Use ViTCA instead of the strongest listed CA baseline (UNetCA) | Denoising autoencoding · CIFAR10 | PSNR (dB): 25.49 -> 25.61 | Helped |
| Use ViTCA instead of the strongest listed CA baseline (UNetCA) | Denoising autoencoding · Tiny ImageNet | PSNR (dB): 21.78 -> 21.58 | Hurt |
| Increase attention heads from 4 to 32 (ablation) | ViTCA · CelebA | PSNR (dB): 26.53 -> 27.01 | Helped |
| Increase encoder depth from 1 to 2 (ablation) | ViTCA · CelebA | PSNR (dB): 26.53 -> 10.82 | Hurt |
| Remove explicit positional information (ablation) | ViTCA · CelebA | PSNR (dB): 26.53 -> 26.28 | No clear effect |
| Reduce cell hidden size from 32 to 8 (ablation) | ViTCA · CelebA | PSNR (dB): 26.53 -> 25.61 | Hurt |
| Increase embedding size from 128 to 512 (ablation) | ViTCA · CelebA | PSNR (dB): 26.05 -> 19.93 | Hurt |
| Increase attention neighborhood from 3x3 to 7x7 (ablation) | ViTCA · FashionMNIST | PSNR (dB): 23.25 -> 21.65 | Hurt |
Closest papers by meaning
- Attention-Only Transformers via Unrolled Subspace Denoising (ICML 2025 · similarity 0.85)
- Learning Linear Attention in Polynomial Time (NeurIPS 2025 · similarity 0.84)
- 2D Spatial Reasoning with Adaptive Neural Cellular Automata (NeurIPS 2026 · similarity 0.84)
- An Evolved Universal Transformer Memory (ICLR 2025 · similarity 0.84)
- Token Statistics Transformer: Linear-Time Attention via Variational Rate Reduction (ICLR 2025 · similarity 0.84)
- Clustering in Causal Attention Masking (NeurIPS 2024 · similarity 0.83)
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.