Research map / Probabilistic and scientific modeling / Dynamical systems
CREIMBO: Cross-Regional Ensemble Interactions in Multi-view Brain Observations
Read the paperCheck your idea against 108,967 papers
AI-extracted summary. Written by AI from the paper's full text. Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Proposes CREIMBO, a graph-regularized dictionary-learning plus sparse time-varying linear dynamics model that fits non-simultaneous multi-session, multi-region neural recordings into shared sub-circuits.
Problem
Neural recordings from different sessions capture different, non-overlapping neuron subsets across brain areas, and existing methods either merge sessions only crudely, assume stationary dynamics, or give uninterpretable latent dynamics; a unified interpretable model of non-stationary multi-regional
Approach
Model each session's neurons as sparse, same-area functional ensembles (via graph-driven re-weighted L1 dictionary learning) that project from a shared low-dimensional latent space, whose non-linear non-stationary dynamics are a sparse time-varying combination (dLDS-style) of global session-invariant sub-circuits, fit jointly across sessions.
Key results
- On synthetic data, CREIMBO recovers ground-truth observations, ensembles, latent dynamics, and sub-circuits with high correlation across 312 random initializations.
- On human recordings (21 subjects, 41 sessions), CREIMBO identifies 12 sub-circuits, with 10 regions and mostly inter-regional motifs; sub-circuits 10 and 12 were automatically nullified.
- On mouse Neuropixels data, CREIMBO sub-circuit activations predict outcome, early lick, and lick side well above chance (p < 1e-10).
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| jointly identify ensembles and dynamics instead of separating the two steps (ablation) | CREIMBO · Synthetic data | correlation with ground-truth components | Helped |
| fit sessions individually rather than using cross-session information (ablation) | CREIMBO · Synthetic data | correlation with ground-truth components | Hurt |
| remove the localized structure of the ensembles (ablation) | CREIMBO · Synthetic data | correlation with ground-truth components | Hurt |
Surprising findings
- Sub-circuits 10 and 12 collapse to near-zero on human data, which the authors interpret as automatic model selection via sparsity.
- Inferred coefficient transition frequency rises with noise, which the authors suggest could indicate non-dynamical noise.
Limitations
- Relies on linear projections from ensemble to observation space.
- Dictionary learning is computationally demanding.
- Identifiability and generalization to unobserved ensembles depend on assumptions (spectral radius near one, spark condition) that lack theoretical guarantees.
Experimental setup
Datasets: synthetic simulated sessions human high-density electrode recordings (Screening and Sternberg tasks) mouse whole-brain Neuropixels memory-guided movement data
Benchmarks: synthetic ground-truth recovery human high-density electrode recordings mouse Neuropixels task-variable prediction
Metrics: correlation with ground truth -log(MSE) reconstruction error logistic regression prediction accuracy (p-values)
Baselines: SLDS rSLDS mp_rSLDS LFADS Single Session ablation All Regions ablation PCA All Regions ablation
Builds on: dLDS rSLDS mDLAG LFADS
Closest papers by meaning
- Identifying interactions across brain areas while accounting for individual-neuron dynamics with a Transformer-based variational autoencoder (NeurIPS 2025 · similarity 0.87)
- Learning dynamic representations of the functional connectome in neurobiological networks (ICLR 2024 · similarity 0.87)
- Partial observation can induce mechanistic mismatches in data-constrained models of neural dynamics (NeurIPS 2024 · similarity 0.87)
- Meta-Dynamical State Space Models for Integrative Neural Data Analysis (ICLR 2025 · similarity 0.87)
- Accurate Identification of Communication Between Multiple Interacting Neural Populations (ICML 2025 · similarity 0.86)
- Inpainting the Neural Picture: Inferring Unrecorded Brain Area Dynamics from Multi-Animal Datasets (NeurIPS 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.