![[2602.18645] Adaptive Time Series Reasoning via Segment Selection](https://arxiv.org/static/browse/0.3.4/images/arxiv-logo-fb.png){kind=logo}
[2602.18645] Adaptive Time Series Reasoning via Segment Selection
Summary
The paper presents ARTIST, a novel approach to time series reasoning that utilizes adaptive segment selection to improve accuracy in answering natural language questions related to time series data.
Why It Matters
As time series data becomes increasingly prevalent in various fields, effective reasoning over this data is crucial. ARTIST's method of selectively analyzing relevant segments enhances the model's performance, addressing limitations of existing approaches that treat the entire series uniformly. This advancement could lead to more accurate insights in applications ranging from finance to healthcare.
Key Takeaways
- ARTIST improves time series reasoning by focusing on relevant data segments.
- The model employs a controller-reasoner architecture using reinforcement learning.
- It shows a 6.46% accuracy improvement over existing baseline models.
- The approach is particularly effective for rare event localization and multi-segment reasoning tasks.
- Supervised fine-tuning and reinforcement learning enhance performance further.
Computer Science > Machine Learning arXiv:2602.18645 (cs) [Submitted on 20 Feb 2026] Title:Adaptive Time Series Reasoning via Segment Selection Authors:Shvat Messica, Jiawen Zhang, Kevin Li, Theodoros Tsiligkaridis, Marinka Zitnik View a PDF of the paper titled Adaptive Time Series Reasoning via Segment Selection, by Shvat Messica and 4 other authors View PDF Abstract:Time series reasoning tasks often start with a natural language question and require targeted analysis of a time series. Evidence may span the full series or appear in a few short intervals, so the model must decide what to inspect. Most existing approaches encode the entire time series into a fixed representation before inference, regardless of whether or not the entire sequence is relevant. We introduce ARTIST, which formulates time-series reasoning as a sequential decision problem. ARTIST interleaves reasoning with adaptive temporal segment selection. It adopts a controller-reasoner architecture and uses reinforcement learning to train the controller role to select informative segments and the reasoner role to generate segment-conditioned reasoning traces and final answers. During inference, the model actively acquires task-relevant information instead of relying on a static summary of the full sequence. We use a novel hierarchical policy optimization approach for post-training that allows the model to excel in both segment selection and question-answering behavior. We evaluate ARTIST on six time-series re...
