Research map / Time series / Foundation models for time series
Rethinking Multimodal Time-Series Forecasting Evaluation
Haoxin Liu, Yichen Zhou, Rajat Sen, B. Aditya Prakash, Abhimanyu Das
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 calendar/date context to metadata (ablation) | Gemini-2.0-Flash · Context-type ablation | normalized MASE: 0.787 -> 0.670 | Improved |
| add covariate context on top of metadata+date (ablation) | Gemini-2.0-Flash · Context-type ablation | normalized MASE: 0.670 -> 0.674 | No clear effect |
| add event context on top of metadata+date (ablation) | Gemini-2.0-Flash · Context-type ablation | normalized MASE: 0.670 -> 0.674 | No clear effect |
| add both events and covariates to metadata+date (ablation) | Gemini-2.0-Flash · Context-type ablation | normalized MASE: 0.670 -> 0.650 | Improved |
| replace Time-MMD text context with TimesX-generated events (ablation) | Gemini-2.0-Flash · Time-MMD (9 datasets) | MASE: 0.906 -> 0.840 | Improved |
| evaluate after the LLM knowledge cutoff instead of before | Gemini-2.0-Flash · Search Trend (120 variables) | MASE: 0.514 -> 0.594 | Hurt |
| evaluate after the LLM knowledge cutoff instead of before | DeepSeek-V3 · Search Trend (120 variables) | MASE: 0.606 -> 0.681 | Hurt |
| evaluate after cutoff instead of before | TimesFM-2.5 · Search Trend (120 variables) | MASE: 0.563 -> 0.573 | No clear effect |
| evaluate after cutoff instead of before | Moirai-2.0 · Search Trend (120 variables) | MASE: 0.691 -> 0.696 | No clear effect |
| average TimesFM-2.5 and Gemini-2.0-Flash forecasts | ensemble · TimesX (post-2024-07-01 horizons) | normalized MASE: 0.645 -> 0.619 | Improved |
| have Gemini revise TimesFM-2.5 forecast via code (CodeRev) | TimesFM-2.5 · TimesX (post-2024-07-01 horizons) | normalized MASE: 0.645 -> 0.713 | Hurt |
| have Gemini revise TimesFM-2.5 forecast via text (TextRev) | TimesFM-2.5 · TimesX (post-2024-07-01 horizons) | normalized MASE: 0.645 -> 0.653 | No clear effect |
Closest papers by meaning
- Context is Key: A Benchmark for Forecasting with Essential Textual Information (ICML 2025 · similarity 0.91)
- Rethinking Post-Training Recipes for Multimodal Time-Series Forecasting (NeurIPS 2026 · similarity 0.90)
- Time-MMD: Multi-Domain Multimodal Dataset for Time Series Analysis (NeurIPS 2024 · similarity 0.90)
- Time-IMM: A Dataset and Benchmark for Irregular Multimodal Multivariate Time Series (NeurIPS 2025 · similarity 0.90)
- Time-VLM: Exploring Multimodal Vision-Language Models for Augmented Time Series Forecasting (ICML 2025 · similarity 0.88)
- TimeXL: Explainable Multi-modal Time Series Prediction with LLM-in-the-Loop (NeurIPS 2025 · similarity 0.88)
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.
