Research map / Learning theory and algorithms / Matrix and tensor methods
Learning Grid Cells as Vector Representation of Self-Position Coupled with Matrix Representation of Self-Motion
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 |
|---|---|---|---|
| Add decoding-encoding at each step (ablation) | Proposed model · Path integral with Gaussian noise (1s) | MSE: 6.578 -> 1.687 | Helped |
| Add decoding-encoding at each step (ablation) | Proposed model · Path planning with Gaussian noise (1s) | success rate: 0.503 -> 0.928 | Helped |
| Add decoding-encoding at each step (ablation) | Proposed model · Path integral with 70% dropout | MSE: 19.611 -> 2.837 | Helped |
| Add decoding-encoding at each step (ablation) | Proposed model · Path planning with 70% dropout | success rate: 0.067 -> 0.810 | Helped |
| Train with multi-step rather than single-step motion loss (ablation) | Proposed model · Path integral | MSE | Helped |
| Use Gaussian adjacency with block size >= 3, or exponential adjacency with block size >= 4, rather than a single unblocked motion representation (ablation) | Proposed model · Path integral | MSE | No clear effect |
| Remove local isometry while retaining motion and global adjacency losses (ablation) | Proposed model · Grid-pattern emergence | grid-like pattern emergence | Hurt |
| Use only global adjacency loss (ablation) | Proposed model · Grid-pattern emergence | grid-like pattern emergence | Hurt |
| Use only local isometry loss (ablation) | Proposed model · Grid-pattern emergence | grid-like pattern emergence | Hurt |
| Remove the quadratic motion-matrix parameterization and use global adjacency with motion loss (ablation) | Proposed model · Grid-pattern emergence | grid-like pattern emergence | Hurt |
Closest papers by meaning
- On Path Integration of Grid Cells: Group Representation and Isotropic Scaling (NeurIPS 2021 · similarity 0.89)
- On Conformal Isometry of Grid Cells: Learning Distance-Preserving Position Embedding (ICLR 2025 · similarity 0.89)
- Flexible mapping of abstract domains by grid cells via self-supervised extraction and projection of generalized velocity signals (NeurIPS 2024 · similarity 0.82)
- Place Cells as Multi-Scale Position Embeddings: Random Walk Transition Kernels for Path Planning (NeurIPS 2025 · similarity 0.81)
- Not so griddy: Internal representations of RNNs path integrating more than one agent (NeurIPS 2024 · similarity 0.80)
- A simple model of co-emergence of grid and place fields (NeurIPS 2026 · 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.