Large language models learn time series perception before reasoning
Perceive, Solve, Reason: Progressive Capability Formation in Time Series-LLMs
Computation and Language
Summary
Understanding time series data like heartbeats or stock prices is harder for language AI than images or sounds. The authors created a big dataset to help AI models learn three key skills in order: noticing details, solving problems, and reasoning about time series. They found that giving the AI a special tool to process the raw data, rather than turning it into words first, improves accuracy a lot. Teaching these skills step-by-step works better than mixing them all at once, and this approach beats bigger models that do not follow their recipe. They suggest treating time series data as a unique type of input for AI, not just text.
What this means in practice
- •For financial analysts: Develop AI tools that better recognize and interpret complex financial time series by following progressive training of perception, solving, and reasoning.
- •For industrial process engineers: Create improved monitoring systems that analyze sensor time series data more accurately by using dedicated encoders and staged learning methods.
Authors
Xin Qiu, Junlong Tong, Wei Zhang, Xiaoyu Shen
Abstract
Large language models (LLMs) can now perceive, describe, and reason about images and audio, yet time series understanding remains underestimated: most work uses LLMs as forecasters and classifiers, reasoning studies are few and narrow, and no suite trains and evaluates the full range of capabilities. In this paper, we build the first such suite, with 337K training and 44K evaluation samples across three progressive capabilities, Perceive, Solve, and Reason, and use it to ask how time series should enter an LLM that must master all three, and in what order. We find that a dedicated encoder is essential: with one, a 4B model surpasses GPT-4o on every perception attribute we test, whereas textualizing the series costs 8--10 points in reasoning accuracy. The capabilities then build on one another: learning them in order beats mixing the same data and rivals reasoning-only training with over three times the reasoning data, but only when an encoder is used. Analyses further reveal that capabilities are transferred through distinct network depths rather than simply stacked, and that a stage converged on its own benchmark is not necessarily ready for the next. Distilled into a simple recipe, these findings yield 4B TS-LLMs that, without any target-domain data, outperform four recent 7B TS-LLMs on seven of eight out-of-domain metrics, by up to 11 points. We therefore advocate treating time series as a first-class modality for LLMs, read through dedicated encoders and taught to perceive before they reason.