Research map / Multi-agent systems and games / Human-AI and goal-conditioned
Symmetry-Based Disentangled Representation Learning requires Interaction with Environments
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 transitions with actions rather than unordered still observations | Learning a symmetry-based representation · Worlds with the same state set and symmetries but different symmetry actions | ability to identify the symmetry action | Helped |
| Jointly learn the representation and action matrices using transition prediction | Forward-VAE · 2D cyclic navigation environment | representation quality | Helped |
| Learn action matrices jointly with the representation | Forward-VAE · 2D cyclic navigation environment | matrix mean squared difference: order of 10^{-4} | Helped |
| Use a 4-dimensional CCI-VAE instead of Forward-VAE (ablation) | Learning the linear symmetry-based representation · 2D cyclic navigation environment | representation quality | Hurt |
| Remove separate latent subspaces for the x and y symmetries (ablation) | Forward-VAE · 2D cyclic navigation environment | disentanglement | Hurt |
| Use an SB-disentangled representation rather than a non-disentangled representation | Inverse model prediction · Limited training data and high classifier capacity | 10-fold cross-validation accuracy | Helped |
| Use an LSB-disentangled representation rather than the alternative representations | Inverse model prediction · Large training set and limited classifier capacity | 10-fold cross-validation accuracy | Helped |
| Use a disentangled representation rather than a non-disentangled one | Inverse model prediction · Sufficient data and classifier capacity | 10-fold cross-validation accuracy | No clear effect |
Closest papers by meaning
- Disentangled representation learning through unsupervised symmetry group discovery (ICLR 2026 · similarity 0.94)
- Expected Return Symmetries (ICLR 2025 · similarity 0.88)
- Latent Mixture of Symmetries for Sample-Efficient Dynamic Learning (NeurIPS 2025 · similarity 0.86)
- Disentangling Representations through Multi-task Learning (ICLR 2025 · similarity 0.85)
- Disentangled Representation Learning in Non-Markovian Causal Systems (NeurIPS 2024 · similarity 0.85)
- A Generative Model of Symmetry Transformations (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.