Research map / Probabilistic and scientific modeling / Dynamical systems
Learning Robust Dynamics through Variational Sparse Gating
Arnav Kumar Jain, Shiva Kanth Sujit, Shruti Joshi, Vincent Michalski, Danijar Hafner, Samira Ebrahimi Kahou
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 GRU recurrent update with variational sparse gating (VSG) | DreamerV2 · BringBackShapes Basic, 0 distractors | first-visit time: 500.25 -> 276.20 | Improved |
| replace GRU recurrent update with VSG | DreamerV2 · BringBackShapes Basic, 0 distractors | visited objects not scored (%): 14.67 -> 1.38 | Improved |
| replace GRU recurrent update with VSG | DreamerV2 · BringBackShapes Basic, 2 distractors | visited objects not scored (%): 35.90 -> 9.67 | Improved |
| replace GRU recurrent update with VSG | DreamerV2 · BringBackShapes Medium, 2 distractors | objects not visited (%): 40.80 -> 16.4 | Improved |
| replace RSSM with purely stochastic single-path SVSG | DreamerV2 · BringBackShapes Basic, 0 distractors | episode length: 2503.55 -> 2196.37 | Improved |
| replace RSSM with SVSG | DreamerV2 · BringBackShapes Medium, 2 distractors | visited objects not scored (%): 45.48 -> 51.42 | Hurt |
| replace RSSM with SVSG | DreamerV2 · BringBackShapes Medium, 0 distractors | visited objects not scored (%): 33.68 -> 38.53 | Hurt |
| use VSG instead of DreamerV2 at 1M steps | DreamerV2 · BringBackShapes Basic, no distractors | mean score: 4.9 | Improved |
| train longer to 2.5M steps | DreamerV2 · BringBackShapes Basic, no distractors | mean score | No clear effect |
| increase arena size (Small to Large) with VSG | VSG · BringBackShapes | score | Hurt |
| use SVSG on larger arenas | SVSG · BringBackShapes larger arenas | score | Hurt |
| add distractor objects (0 to 4) | VSG, SVSG, DreamerV2 · BringBackShapes | score | Hurt |
Closest papers by meaning
- Learning interacting dynamical systems with latent Gaussian process ODEs (NeurIPS 2022 · similarity 0.88)
- SPARTAN: A Sparse Transformer World Model Attending to What Matters (NeurIPS 2025 · similarity 0.86)
- Zero-shot World Models via Search in Memory (NeurIPS 2025 · similarity 0.85)
- Sparse Diffusion Autoencoder for Test-time Adapting Prediction of Complex Systems (NeurIPS 2025 · similarity 0.85)
- Imagine Beyond ! Distributionally Robust Autoencoding for State Space Coverage in Online Reinforcement Learning (NeurIPS 2025 · similarity 0.85)
- Simplifying Latent Dynamics with Softly State-Invariant World Models (NeurIPS 2024 · similarity 0.85)
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.
