Research map / Time series / Foundation models for time series
Adaptive Time Series Reasoning via Segment Selection
Shvat Messica, Jiawen Zhang, Kevin Li, Theodoros Tsiligkaridis, Marinka Zitnik
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 RL stage after SFT (ablation) | ARTIST · Average over six datasets | accuracy (%): 63.61 -> 69.26 | Improved |
| add adaptive segment selection controller (ablation) | ARTIST · ECG-QA + RCW average | accuracy (%): 9.30% average accuracy drop without controller | Improved |
| add joint reasoner training (unfrozen reasoner) in RL (ablation) | ARTIST · ECG-QA + RCW average | accuracy (%): 8.94% decrease with controller-only RL | Improved |
| add reliability reward for controller (ablation) | ARTIST · ECG-QA + RCW average | accuracy (%): 73.41 -> 51.97 when removed (21.44% drop) | Improved |
| add trajectory-level (hierarchical) controller objective instead of myopic (ablation) | ARTIST · ECG-QA + RCW average | accuracy (%): 12.28% drop with myopic objective | Improved |
| add variance-guided sampling for reasoner group selection (ablation) | ARTIST · ECG-QA + RCW average | accuracy (%): 73.41 -> 70.44 when removed (2.97%) | Improved |
| use VLM backbone with plotted time series instead of tokenized input | ARTIST (SFT only) · Sleep-QA | accuracy (%): 28.13 -> 65.00 | Improved |
| use ARTIST segment selection instead of PixelReasoner on Qwen3-4B | Qwen3-4B · ECG-QA | accuracy (%): 60.15 -> 69.81 | Improved |
| use ARTIST instead of strongest baseline | ARTIST · Average over six datasets | accuracy (%): +6.46 | Improved |
| add more of the time series (near-complete usage) | ARTIST · Sleep-QA and TRQA | accuracy (%) | Hurt |
| add RL on TimeMaster vs ARTIST on Sleep-QA | TimeMaster · Sleep-QA | accuracy (%): TimeMaster+RL 72.55 vs ARTIST 36.63 | Mixed |
| add statistics to prompt | GPT-5 · ETI | accuracy (%): 29.50 -> 63.54 | Improved |
Closest papers by meaning
- Large Pre-trained time series models for cross-domain Time series analysis tasks (NeurIPS 2024 · similarity 0.85)
- ITFormer: Bridging Time Series and Natural Language for Multi-Modal QA with Large-Scale Multitask Dataset (ICML 2025 · similarity 0.85)
- Tracing the Traces: Latent Temporal Signals for Efficient and Accurate Reasoning (ICLR 2026 · similarity 0.84)
- ReXTime: A Benchmark Suite for Reasoning-Across-Time in Videos (NeurIPS 2024 · similarity 0.84)
- TimeClaw: A Time-Series AI Agent with Exploratory Execution Learning (NeurIPS 2026 · similarity 0.84)
- Time-LLM: Time Series Forecasting by Reprogramming Large Language Models (ICLR 2024 · similarity 0.84)
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.
