Research map / Time series / Foundation models for time series
StarEmbed: Benchmarking Time Series Foundation Models on Astronomical Observations of Variable Stars
Weijian Li, Hong-Yu Chen, Nabeel Rehemtulla, Ved G. Shah, Dongho Kim, Dennis Wu, Qinjie Lin, Adam A. Miller, Han Liu
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 |
|---|---|---|---|
| use Chronos-Bolt embeddings with multi-class isolation forest | OOD detection · StarEmbed OOD | OOD purity top 10%: 0.259 -> 0.519 vs hand-crafted | Improved |
| swap isolation forest for Local Outlier Factor (ablation) | Chronos OOD detection · StarEmbed OOD top 10% | OOD purity: 0.149 -> 0.992 | Improved |
| fine-tune Chronos with classification head instead of forecasting objective (ablation) | Chronos-tiny · StarEmbed classification (MLP) | macro F1: 0.627 -> 0.659 | Improved |
| fine-tune Chronos with forecasting objective (ablation) | Chronos-tiny · StarEmbed classification (MLP) | macro F1: 0.643 -> 0.627 | Hurt |
| use LoRA instead of full fine-tuning with linear head (ablation) | Chronos · StarEmbed classification | macro F1: 0.661 -> 0.622 | Hurt |
| use LoRA instead of full fine-tuning with MLP head (ablation) | Chronos-tiny · StarEmbed classification | macro F1: 0.659 -> 0.661 | No clear effect |
| tune LayerNorm only instead of LoRA with MLP head (ablation) | Chronos-tiny · StarEmbed classification | macro F1: 0.661 -> 0.648 | Hurt |
| replace patch-index RoPE with true timestamps (ablation) | Moirai · synthetic sinusoids | R2: 0.016 -> 0.188 | Improved |
| append time gaps as covariate instead of RoPE timestamps (ablation) | Moirai · synthetic sinusoids | R2: 0.188 -> 0.044 | Hurt |
| scale Chronos from Tiny to Large (ablation) | Chronos · StarEmbed k-NN | macro F1: 0.672 -> 0.646 | No clear effect |
| scale Moirai from Small to Large (ablation) | Moirai · StarEmbed k-NN | macro F1: 0.554 -> 0.612 | Improved |
| full fine-tune Mantis models (ablation) | MantisV2 · StarEmbed classification (MLP) | macro F1: 0.570 -> 0.638 | Improved |
Closest papers by meaning
- Probabilistic Data-Driven Modelling of Astrophysical Transients: The Neural Process Family for Ultrafast and Class-Agnostic Light Curve Reconstruction (NeurIPS 2026 · similarity 0.86)
- AstroVisBench: A Code Benchmark for Scientific Computing and Visualization in Astronomy (NeurIPS 2025 · similarity 0.83)
- A New Statistical Model of Star Speckles for Learning to Detect and Characterize Exoplanets in Direct Imaging Observations (CVPR 2025 · similarity 0.83)
- The Multimodal Universe: Enabling Large-Scale Machine Learning with 100 TB of Astronomical Scientific Data (NeurIPS 2024 · similarity 0.83)
- CosmoBench: A Multiscale, Multiview, Multitask Cosmology Benchmark for Geometric Deep Learning (NeurIPS 2025 · similarity 0.82)
- VESTA: Visual Exploration with Statistical Tool Agents (NeurIPS 2026 · similarity 0.82)
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.
