Research map / Architectures and efficiency / Attention and transformers
Facing Off World Model Backbones: RNNs, Transformers, and S4
Fei Deng, Junyeong Park, Sungjin Ahn
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 |
|---|---|---|---|
| remove MLP in S4 blocks (ablation) | S4WM · Four Rooms | generation MSE: 44.0 -> 68.6 | Hurt |
| remove MLP in S4 blocks (ablation) | S4WM · Ten Rooms | generation MSE: 224.4 -> 277.0 | Hurt |
| use full posterior (ablation) | S4WM · Four Rooms | generation MSE: 44.0 -> 38.0 | Improved |
| use full posterior (ablation) | S4WM · Ten Rooms | generation MSE: 224.4 -> 171.1 | Improved |
| use full posterior (ablation) | S4WM · Ten Rooms | reconstruction MSE: 1.8 -> 2.0 | Hurt |
| increase TSSM-XL cache length from 128 to 512 (ablation) | TSSM-XL · Four Rooms | generation MSE: 224.4 -> 38.4 | Improved |
| increase TSSM-XL cache length from 128 to 512 (ablation) | TSSM-XL · Ten Rooms | generation MSE: 360.4 -> 277.6 | Improved |
| increase TSSM-XL cache length from 128 to 256 (ablation) | TSSM-XL · Four Rooms | generation MSE: 224.4 -> 124.4 | Improved |
| replace S4 with S5 in the world model (ablation) | S4WM · Ten Rooms | generation MSE: 224.4 -> 159.2 | Improved |
| replace S4 with S5 in the world model (ablation) | S4WM · Two Rooms | generation MSE: 27.3 -> 10.6 | Improved |
| replace S4 with S5 in the world model (ablation) | S4WM · Four Rooms | generation MSE: 44.0 -> 19.2 | Improved |
| use S4WM instead of RSSM-TBTT backbone (ablation) | world model · Four Rooms | generation MSE: 219.4 -> 44.0 | Improved |
Closest papers by meaning
- Mastering Memory Tasks with World Models (ICLR 2024 · similarity 0.89)
- Zero-shot World Models via Search in Memory (NeurIPS 2025 · similarity 0.88)
- Learning Transformer-based World Models with Contrastive Predictive Coding (ICLR 2025 · similarity 0.88)
- EDELINE: Enhancing Memory in Diffusion-based World Models via Linear-Time Sequence Modeling (NeurIPS 2025 · similarity 0.87)
- Learning Latent Dynamic Robust Representations for World Models (ICML 2024 · similarity 0.87)
- RLVR-World: Training World Models with Reinforcement Learning (NeurIPS 2025 · similarity 0.87)
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.
