Research map / Probabilistic and scientific modeling / Variational and generative sampling
[Re] Variational Neural Cellular Automata
Albert Sund Aillet, Simon Sondén
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 |
|---|---|---|---|
| Use doubling NCA decoder instead of non-doubling variant (ablation) | VNCA · Binarized MNIST | log p(x) (nats): -90.97 -> -84.23 | Improved |
| Use doubling NCA decoder instead of non-doubling variant (ablation) | VNCA · CelebA | bits per dimension: 5.03 -> 4.42 | Improved |
| Increase KL weight beta from 1 to 100 (ablation) | VNCA · CelebA | bits per dimension: 4.42 -> 5.40 | Mixed |
| Pool and damage recovery training with non-doubling VNCA | VNCA · MNIST damage recovery | recovery from damage | Improved |
| Replace VNCA decoder with deep convolutional VAE decoder (ablation) | VNCA · CelebA damage during growth | damage recovery | No clear effect |
| Deeper (up to 8 residual layers) and wider (512) update network (ablation) | VNCA · generative modeling | likelihood | No clear effect |
| Latent size 512 and NCA hidden state 512 (ablation) | VNCA · generative modeling | likelihood | No clear effect |
| Use L=4 logistic mixtures instead of L=1 (ablation) | VNCA · generative modeling | likelihood | Hurt |
| Replace 1x1 convolutions with 3x3 (ablation) | VNCA · generative modeling | likelihood | No clear effect |
| Separate update network per scale (K networks) (ablation) | VNCA · generative modeling | likelihood | No clear effect |
| Increase batch size (batch 4 vs 32) (ablation) | VNCA · training | trainability | Improved |
| Use GRU update, gated residual update, or half-frozen hidden state (ablation) | VNCA · generative modeling | likelihood | No clear effect |
Closest papers by meaning
- Variational Neural Cellular Automata (ICLR 2022 · similarity 0.88)
- Probing the Inductive Bias of Neural Networks through Learning Random Cellular Automata (ICML 2026 · similarity 0.84)
- 2D Spatial Reasoning with Adaptive Neural Cellular Automata (NeurIPS 2026 · similarity 0.81)
- Brain-like Variational Inference (NeurIPS 2025 · similarity 0.81)
- CAX: Cellular Automata Accelerated in JAX (ICLR 2025 · similarity 0.80)
- eXponential FAmily Dynamical Systems (XFADS): Large-scale nonlinear Gaussian state-space modeling (NeurIPS 2024 · similarity 0.80)
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.
