Research map / Probabilistic and scientific modeling / Dynamical systems
A Factorized Low-Rank RNN Framework for Uncovering Independent Neural Latent Dynamics and Connectivity
Chengrui Li, Yunmiao Wang, Yule Wang, Weihan Li, Dieter Jaeger, Anqi Wu
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 |
|---|---|---|---|
| add group-wise independence (partial correlation penalty) to the low-rank RNN | lrRNN · Synthetic Lorenz+Thomas latents | latent partial correlation: 0.86 -> 0.13 | Improved |
| train factorization end-to-end instead of post-hoc ICA on lrRNN latents (ablation) | FacRNN · Synthetic data | latent factorization | Improved |
| add factorization to low-rank RNN | lrRNN/LINT · Synthetic data | reconstruction R2: approximately 0.85 for all methods | No clear effect |
| add factorization with two rank-1 groups | lrRNN · Monkey M1 (K=2) | alignment R2: 0.5 -> 0.65 | Improved |
| use two rank-2 groups at K=4 (ablation) | FacRNN · Monkey M1 (K=4) | alignment R2: about 0.8 | Improved |
| use 4 rank-1 groups (full independence) instead of 2 rank-2 groups (ablation) | FacRNN · Monkey M1 (K=4) | alignment R2 | Hurt |
| replace linear encoder/decoder with MLP (ablation) | FacRNN · Monkey M1 (K=2) | alignment R2 | No clear effect |
| use sophisticated SMC inference without direction labels (ablation) | lrRNN · Monkey M1 (K=4) | alignment R2 | Hurt |
| increase total latent components K up to six (ablation) | FacRNN (G=K, H=1) · Mouse voltage imaging | reconstruction R2 | Improved |
| vary number of groups G at fixed K=12 (ablation) | FacRNN · Mouse voltage imaging | reconstruction R2 | No clear effect |
| use 6 rank-2 groups instead of 6 rank-1 groups (ablation) | FacRNN · Mouse voltage imaging | between-group total correlation: 0.159 -> 0.110 | Improved |
Closest papers by meaning
- Inferring stochastic low-rank recurrent neural networks from neural data (NeurIPS 2024 · similarity 0.89)
- Learning Reveals Invisible Structure in Low-Rank RNNs (NeurIPS 2026 · similarity 0.88)
- Leveraging Low-Rank and Sparse Recurrent Connectivity for Robust Closed-Loop Control (ICLR 2024 · similarity 0.87)
- Mechanistic Interpretability of RNNs emulating Hidden Markov Models (NeurIPS 2025 · similarity 0.87)
- Modeling Latent Neural Dynamics with Gaussian Process Switching Linear Dynamical Systems (NeurIPS 2024 · similarity 0.86)
- BLEND: Behavior-guided Neural Population Dynamics Modeling via Privileged Knowledge Distillation (ICLR 2025 · 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.
