Adaptive Time Series Reasoning via Segment Selection

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 relevance. 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 reasoning benchmarks and compare it with large language models, vision-language models, and prior time-series reasoning systems. ARTIST improves average accuracy by 6.46 absolute percentage points over the strongest baseline. The largest gains appear on rare event localization and multi-segment reasoning tasks. Supervised fine-tuning improves performance, and reinforcement learning provides additional gains by optimizing question-adaptive segment selection, showing that selective data use drives effective time-series reasoning.

When a doctor investigates whether a treatment is working for a patient, or an analyst studies what caused a market to shift, they don't stare at the entire history all at once. They scan the data, focus on key areas, analyze them, draw initial insights, and then decide whether they need to look elsewhere or compare against another period to confirm what they're seeing. Today's AI systems work very differently: they try to digest the whole sequence in one pass, which can bury the important clues in irrelevant noise. We built ARTIST, an AI system that reads time-based data the way a careful expert would. ARTIST plays two roles that work together: a "controller" that identifies the specific slices of data essential for answering the question, and a "reasoner" that interprets what the controller provides and produces an answer. After each round, the controller reviews the reasoner's answer and decides whether to provide more data or stop. We trained these two roles to cooperate using a learning method that rewards them for working as a team rather than rushing to a guess. Across six tests spanning medicine, finance, and the environment, ARTIST answered questions more accurately than leading AI systems while examining only a fraction of the data - showing that knowing where to look matters as much as looking carefully.