Research map / Time series / Causal and generative time series
Panda: A pretrained forecast model for chaotic dynamics
Jeffrey B. Lai, Anthony Bao, William Gilpin
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 channel attention (multivariate) to the patch transformer (ablation) | Panda · held-out chaotic systems | sMAPE | Improved |
| add masked-patch (MLM) pretraining before forecast training (ablation) | Panda · held-out chaotic systems, short horizon | sMAPE | Improved |
| add MLM pretraining (ablation) | Panda · autoregressive rollout, long horizon | sMAPE | Hurt |
| add dynamics embedding (polynomial and random Fourier features) without MLM (ablation) | Panda · held-out chaotic systems | sMAPE | Improved |
| add dynamics embedding on top of MLM pretraining (ablation) | Panda with MLM · held-out chaotic systems | sMAPE | Hurt |
| add dynamics embedding with polynomial features (ablation) | Panda · autoregressive rollout, long horizon | sMAPE | Improved |
| finetune Chronos 20M on the chaotic systems dataset | Chronos 20M · long-term forecast KL divergence, L_pred=512 | KL divergence: 5.99 -> 4.72 | Improved |
| increase number of unique systems at constant total timepoints (ablation) | Panda · held-out chaotic systems | zero-shot forecast error | Improved |
| use univariate (channel-independent) model instead of channel attention (ablation) | Panda · circuit dataset with increasing coupling | forecast error | Hurt |
| extend forecast horizon far beyond training horizon | Panda vs Chronos 20M SFT · long-term KL divergence | KL divergence: 3.93 -> 6.39 (Panda); 4.72 -> 6.05 (Chronos SFT) | Mixed |
Closest papers by meaning
- Zero-shot forecasting of chaotic systems (ICLR 2025 · similarity 0.90)
- Chaos Meets Attention: Transformers for Large-Scale Dynamical Prediction (ICML 2025 · similarity 0.88)
- When are dynamical systems learned from time series data statistically accurate? (NeurIPS 2024 · similarity 0.88)
- Attractor Memory for Long-Term Time Series Forecasting: A Chaos Perspective (NeurIPS 2024 · similarity 0.87)
- Almost-Linear RNNs Yield Highly Interpretable Symbolic Codes in Dynamical Systems Reconstruction (NeurIPS 2024 · similarity 0.86)
- Stable Long-Horizon PDE Forecasting via Latent Structured Spectral Propagators (NeurIPS 2026 · 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.
