Research map / Time series / Causal and generative time series
Reconstructing Nonlinear Dynamical Systems from Multi-Modal Time Series
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 categorical task labels to BOLD observations | mmPLRNN (EM) · fMRI working-memory task | attractor geometry KL divergence: ≈ 2.56 ± 0.03 -> ≈ 1.09 ± 0.03 | Helped |
| Add categorical task labels to BOLD observations | mmPLRNN · fMRI working-memory task | n-step ahead prediction MSE: up to 26% | Helped |
| Add categorical orthant labels to noisy Gaussian observations | PLRNN · Noisy Lorenz, T=1000 | normalized attractor geometry KL divergence | Helped |
| Add categorical orthant labels when one continuous Lorenz variable is missing | PLRNN · Lorenz with x2 omitted | attractor geometry KL divergence | Helped |
| Add categorical labels at low continuous-observation noise (ablation) | PLRNN trained by VI · Noisy Lorenz, T=2500, 5% observation noise | attractor geometry KL divergence | No clear effect |
| Train VI on longer sequences (ablation) | mmPLRNN (VI) · Noisy Lorenz, T=2500 | normalized attractor geometry KL divergence: D̄_KL ≈ 0.49 (n=40) | Helped |
| Reset simulated states to inferred values at instruction onsets (ablation) | mmPLRNN (EM) · fMRI working-memory task | task-label alignment | Helped |
| Reset simulated states at arbitrary, similarly spaced time points instead of instruction onsets (ablation) | mmPLRNN (EM) · fMRI working-memory task | task-label alignment | Hurt |
| Replace all simulated latent states with inferred values (ablation) | mmPLRNN (EM) · fMRI working-memory task | task-label alignment | No clear effect |
| Run free simulation without task-related resets (ablation) | mmPLRNN (EM) · fMRI working-memory task | task-label F1 across thirds: F1≈0.38 -> F1≈0.17 and F1≈0.16 | Hurt |
Closest papers by meaning
- Integrating Multimodal Data for Joint Generative Modeling of Complex Dynamics (ICML 2024 · similarity 0.91)
- Learning Interpretable Hierarchical Dynamical Systems Models from Time Series Data (ICLR 2025 · similarity 0.91)
- A scalable generative model for dynamical system reconstruction from neuroimaging data (NeurIPS 2024 · similarity 0.90)
- Parallel-in-Time Training of Recurrent Neural Networks for Dynamical Systems Reconstruction (NeurIPS 2026 · similarity 0.89)
- Probabilistic Decomposed Linear Dynamical Systems for Robust Discovery of Latent Neural Dynamics (NeurIPS 2024 · similarity 0.89)
- Almost-Linear RNNs Yield Highly Interpretable Symbolic Codes in Dynamical Systems Reconstruction (NeurIPS 2024 · similarity 0.89)
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.