SciTS: Scientific Time Series Understanding and Generation with LLMs

2025/09/26 by Wen Wu, Ziyang Zhang, Wu, Wen +24
Computer Science · Decision Sciences · #Artificial Intelligence (cs.AI) #FOS: Computer and information sciences #Machine Learning (cs.LG) #Scientific Computing and Data Management #Time Series Analysis and Forecasting

paper · pdf · doi:10.48550/arxiv.2510.03255

openalex publication_date 2025/09/26 · openalex created_date 2025/10/09 · openalex updated_date 2026/07/28

Abstract

The scientific reasoning ability of large language models (LLMs) has recently attracted significant attention. Time series, as a fundamental modality in scientific data, presents unique challenges that are often overlooked in current multimodal LLMs, which either encode numerical sequences as text or convert them into images. Such approaches may be insufficient for comprehensive scientific time series understanding and generation. Existing unified time series models typically specialise in either forecasting or analysis, and their effectiveness on non-periodic, heterogeneous scientific signals remains unclear. To address these gaps, we introduce SciTS, a benchmark spanning 12 scientific domains and 43 tasks, with over 50k+ instances, both univariate and multivariate signals ranging from 10⁰ to 10⁷ in length and up to 10~MHz in frequency. We benchmark 17 models, including text-only LLMs, multimodal LLMs, and unified time series models, and find that general-purpose LLMs exhibit stronger generalisability than specialised time series models, while representing time series as text or images limits their performance due to excessively long sequences and loss of numerical precision, respectively. We then introduce TimeOmni, a framework that equips LLMs with the ability to understand and generate time series while remaining compatible with general-purpose LLM training. This work fills a gap in both dedicated benchmarks and modelling frameworks for scientific time series, paving the way for LLMs to understand and generate complex temporal scientific data.

Citations

• InternVL3.5: Advancing Open-Source Multimodal Models in Versatility, Reasoning, and Efficiency
• Scientists' First Exam: Probing Cognitive Abilities of MLLM via Perception, Understanding, and Reasoning
• FinMultiTime: A Four-Modal Bilingual Dataset for Financial Time-Series Analysis
• The iNaturalist Sounds Dataset
• Qwen3 Technical Report
• T2S: High-resolution Time Series Generation with Text-to-Series Diffusion Models
• ChartQAPro: A More Diverse and Challenging Benchmark for Chart Question Answering
• MTBench: A Multimodal Time Series Benchmark for Temporal Reasoning and Question Answering
• S4M: S4 for multivariate time series forecasting with Missing values
• Time-MQA: Time Series Multi-Task Question Answering with Context Enhancement
• Qwen2.5-VL Technical Report
• TimeCAP: Learning to Contextualize, Augment, and Predict Time Series Events with Large Language Model Agents
• TempoGPT: Enhancing Time Series Reasoning via Quantizing Embedding
• Context-Alignment: Activating and Enhancing LLM Capabilities in Time Series
• ChatTS: Aligning Time Series with LLMs via Synthetic Data for Enhanced Understanding and Reasoning
• TimeMixer++: A General Time Series Pattern Machine for Universal Predictive Analysis
• TimeSeriesExam: A time series understanding exam
• Moirai-MoE: Empowering Time Series Foundation Models with Sparse Mixture of Experts
• Timer-XL: Long-Context Transformers for Unified Time Series Forecasting
• Can LLMs Understand Time Series Anomalies?
• From News to Forecast: Integrating Event Analysis in LLM-Based Time Series Forecasting with Reflection
• Time-MoE: Billion-Scale Time Series Foundation Models with Mixture of Experts
• The Llama 3 Herd of Models
• Are Language Models Actually Useful for Time Series Forecasting?
• Time-MMD: Multi-Domain Multimodal Dataset for Time Series Analysis
• TimeMixer: Decomposable Multiscale Mixing for Time Series Forecasting
• Chronos: Learning the Language of Time Series
• UniTS: A Unified Multi-Task Time Series Model
• DoRA: Weight-Decomposed Low-Rank Adaptation
• Unified Training of Universal Time Series Forecasting Transformers
• Multi-stage Learning for Radar Pulse Activity Segmentation
• iTransformer: Inverted Transformers Are Effective for Time Series Forecasting
• Time-LLM: Time Series Forecasting by Reprogramming Large Language Models
• TACTiS-2: Better, Faster, Simpler Attentional Copulas for Multivariate Time Series
• Probabilistic Imputation for Time-series Classification with Missing Data
• Model scale versus domain knowledge in statistical forecasting of chaotic systems
• A Time Series is Worth 64 Words: Long-term Forecasting with Transformers
• Learn to Explain: Multimodal Reasoning via Thought Chains for Science Question Answering
• Out-of-Distribution Representation Learning for Time Series Classification
• ChartQA: A Benchmark for Question Answering about Charts with Visual and Logical Reasoning
• BNAI, NO-TOKEN, and MIND-UNITY: Pillars of a Systemic Revolution in Artificial Intelligence
• Autoformer: Decomposition Transformers with Auto-Correlation for Long-Term Series Forecasting
• MIMII DUE: Sound Dataset for Malfunctioning Industrial Machine Investigation and Inspection with Domain Shifts due to Changes in Operational and Environmental Conditions
• Analysis of a sleep-dependent neuronal feedback loop: the slow-wave microcontinuity of the EEG
• Intern-S1: A Scientific Multimodal Foundation Model

Related

• Large Language Models for Time Series: A Survey
• Large Language models for Time Series Analysis: Techniques, Applications, and Challenges
• Empowering Time Series Analysis with Large Language Models: A Survey
• Forging Time Series with Language: A Large Language Model Approach to Synthetic Data Generation
• Towards Time Series Reasoning with LLMs