Research map / Time series / Long-term forecasting
ChaosNexus: A Foundation Model for ODE-based Chaotic System Forecasting with Hierarchical Multi-scale Awareness
Chang Liu, Bohao Zhao, Jingtao Ding, Yong Li
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 hierarchical patch merging and expansion (ScaleFormer U-Net structure) (ablation) | ChaosNexus · Synthetic chaotic systems | sMAPE@128: 74.161 -> 68.901 | Improved |
| add Mixture-of-Experts layers instead of plain feed-forward (ablation) | ChaosNexus · Synthetic chaotic systems | sMAPE@128: 73.076 -> 68.901 | Improved |
| add MMD-based auxiliary regularization (ablation) | ChaosNexus · Synthetic chaotic systems | sMAPE@512: 110.228 -> 100.293 | Improved |
| add MMD regularization (ablation) | ChaosNexus · Synthetic chaotic systems | D_Lyap: 0.103 -> 0.065 | Improved |
| add wavelet-scattering frequency fingerprint (ablation) | ChaosNexus · Synthetic chaotic systems | sMAPE@128: 72.699 -> 68.901 | Improved |
| pretrain on chaotic systems corpus (fine-tune general time series model) | Chronos-L · Synthetic chaotic systems | D_stsp: 11.731 -> 2.685 | Improved |
| increase model parameters from 2.83M to 52.63M (ablation) | ChaosNexus · Synthetic chaotic systems | sMAPE@128: improved by 49.83% | Improved |
| increase number of training time points per system with fixed system diversity (ablation) | ChaosNexus · Synthetic chaotic systems (zero-shot) | sMAPE | No clear effect |
| increase number of distinct training systems (ablation) | ChaosNexus · Synthetic chaotic systems (zero-shot) | sMAPE | Improved |
| use ChaosNexus instead of Panda as spectral entropy rises (Lorenz-96, F varied) | ChaosNexus · Lorenz-96 varying forcing F | ME_LRw: Panda near 4.5 vs ChaosNexus approx 2.5 | Improved |
| pretrain on synthetic chaotic systems then use zero-shot for forecasting | ChaosNexus · WEATHER-5K temperature, 120-hour horizon | MAE: zero-shot below 1°C vs baselines at least 3°C after fine-tuning | Improved |
| add few-shot fine-tuning on WEATHER-5K | ChaosNexus · WEATHER-5K temperature | MAE | Improved |
Closest papers by meaning
- Zero-shot forecasting of chaotic systems (ICLR 2025 · similarity 0.87)
- Attractor Memory for Long-Term Time Series Forecasting: A Chaos Perspective (NeurIPS 2024 · similarity 0.87)
- Chaos Meets Attention: Transformers for Large-Scale Dynamical Prediction (ICML 2025 · similarity 0.86)
- The Dynamic-Probabilistic Consistency Gap in Chaotic Surrogate Modeling (NeurIPS 2026 · similarity 0.84)
- ChaosBench: A Multi-Channel, Physics-Based Benchmark for Subseasonal-to-Seasonal Climate Prediction (NeurIPS 2024 · similarity 0.84)
- TimeMixer: Decomposable Multiscale Mixing for Time Series Forecasting (ICLR 2024 · similarity 0.83)
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.
