Research map / Probabilistic and scientific modeling / Molecules and proteins
Explaining heterogeneity in medial entorhinal cortex with task-driven neural networks
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 |
|---|---|---|---|
| train network to path integrate (vs untrained architecture) (ablation) | MEC network models · 2D open-field MEC neural predictivity | neural predictivity | Helped |
| use ReLU instead of Linear/Tanh/Sigmoid activation (ablation) | task-optimized RNNs · MEC neural predictivity | neural predictivity | Helped |
| use gated recurrent cells (UGRNN, GRU, LSTM) instead of ungated RNN/SRNN under ReLU (ablation) | task-optimized RNNs · MEC neural predictivity | neural predictivity | Helped |
| use place cell construction objective instead of direct 2D position loss (ablation) | task-optimized RNNs · MEC neural predictivity | neural predictivity | Helped |
| use place cell loss instead of position loss (ablation) | UGRNN ReLU · grid score distribution | KS distance: 0.518 -> 0.039 | Helped |
| train UGRNN ReLU (place cell loss) vs untrained (ablation) | UGRNN ReLU · border score distribution | KS distance: 0.188 -> 0.209 | No clear effect |
| use low-rank NMF Grid Cell model instead of task-optimized network | MEC model · MEC neural predictivity | neural predictivity | Hurt |
| add phantom visual cue locations as input to UGRNN-ReLU-Place Cell (ablation) | UGRNN ReLU Place Cell · 1D track neural predictivity | fraction of noise ceiling: 82% -> noise ceiling | Helped |
| knock out grid or border cells vs matched number of heterogeneous cells (ablation) | UGRNN/LSTM ReLU place cell · path integration performance | performance degradation: at most 1-4% degradation | No clear effect |
| add binary reward input to network without modulating velocity (ablation) | UGRNN ReLU Place Cell · reward+/- MEC data | neural predictivity | No clear effect |
| train with mixture of explore and exploit trajectories (eps = 0.375) vs pure explore (ablation) | UGRNN ReLU Place Cell · reward remapping of MEC units | KS distance: 0.81 -> 0.13 | Helped |
| use Ridge-like linear mapping instead of strict one-to-one mapping (ablation) | inter-animal MEC mapping · 620 mouse MEC cells | inter-animal consistency: around 50% -> > 0.8 | Helped |
Closest papers by meaning
- REMI: Reconstructing Episodic Memory During Internally Driven Path Planning (NeurIPS 2025 · similarity 0.88)
- A Multi-Region Brain Model to Elucidate the Role of Hippocampus in Spatially Embedded Decision-Making (ICML 2025 · similarity 0.87)
- Global Distortions from Local Rewards: Neural Coding Strategies in Path-Integrating Neural Systems (NeurIPS 2024 · similarity 0.87)
- A simple model of co-emergence of grid and place fields (NeurIPS 2026 · similarity 0.87)
- Not so griddy: Internal representations of RNNs path integrating more than one agent (NeurIPS 2024 · similarity 0.86)
- Disentangling the Roles of Distinct Cell Classes with Cell-Type Dynamical Systems (NeurIPS 2024 · similarity 0.86)
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.