Research map / Time series / Foundation models for time series
Aurora: Towards Universal Generative Multimodal Time Series Forecasting
Xingjian Wu, Jianxin Jin, Wanghui Qiu, Peng Chen, Yang Shu, Bin Yang, Chenjuan Guo
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 Modality-Guided Multi-head Self-Attention (ablation) | Aurora · TimeMMD Climate (zero-shot) | MSE: 1.176 -> 0.865 | Improved |
| add Modality-Guided Multi-head Self-Attention (ablation) | Aurora · TimeMMD Economy (zero-shot) | MSE: 0.277 -> 0.033 | Improved |
| add Modality-Guided Multi-head Self-Attention (ablation) | Aurora · TimeMMD Energy (zero-shot) | MSE: 0.268 -> 0.255 | Improved |
| add Prototype-Guided Flow Matching (prototype start instead of Gaussian) (ablation) | Aurora · TimeMMD Traffic (zero-shot) | MSE: 0.273 -> 0.161 | Improved |
| add Prototype-Guided Flow Matching (ablation) | Aurora · TimeMMD Energy (zero-shot) | MSE: 0.257 -> 0.255 | No clear effect |
| add Prototype-Guided Flow Matching (ablation) | Aurora · TimeMMD Social Good (zero-shot) | MSE: 1.425 -> 0.838 | Improved |
| use both modules together (ablation) | Aurora · TimeMMD Climate (zero-shot) | MSE: 1.447 -> 0.865 | Improved |
| use both modules together (ablation) | Aurora · TimeMMD Economy (zero-shot) | MSE: 0.296 -> 0.033 | Improved |
| increase number of flow-matching samples (ablation) | Aurora · ProbTS (avg) | CRPS and NMAE | Improved |
| add text and image modalities with pretraining (Aurora vs unimodal foundation models) | Aurora · TimeMMD (zero-shot) | MSE: average MSE reduction of 27.0% vs Sundial | Improved |
| 10% few-shot fine-tuning of Aurora vs full-shot multimodal baselines | Aurora · TimeMMD | MSE: average MSE reduction of 12.8% vs GPT4MTS | Improved |
| Aurora vs Sundial on Electricity zero-shot | Aurora · TSFM-Bench Electricity | MSE: 0.169 -> 0.178 | Hurt |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 1 of 1 on most shared results.
- Sundial — Sundial: A Family of Highly Capable Time Series Foundation Models (ICML 2025) · better on 8 of 9 shared results
Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.
Closest papers by meaning
- Time-VLM: Exploring Multimodal Vision-Language Models for Augmented Time Series Forecasting (ICML 2025 · similarity 0.90)
- Empowering Time Series Analysis with Large-Scale Multimodal Pretraining (NeurIPS 2026 · similarity 0.90)
- ChatTime: A Unified Multimodal Time Series Foundation Model Bridging Numerical and Textual Data (AAAI 2025 · similarity 0.90)
- Rethinking Post-Training Recipes for Multimodal Time-Series Forecasting (NeurIPS 2026 · similarity 0.90)
- GPT4MTS: Prompt-based Large Language Model for Multimodal Time-series Forecasting (AAAI 2024 · similarity 0.89)
- Time-MMD: Multi-Domain Multimodal Dataset for Time Series Analysis (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.
