arXiv 2406.016382025-03-29cs.LGcs.AIcs.CL

TimeCMA：通过跨模态对齐实现大语言模型赋能的多变量时间序列预测

TimeCMA: Towards LLM-Empowered Multivariate Time Series Forecasting via Cross-Modality Alignment

S-Lab, Nanyang Technological University（南洋理工大学S-Lab）；Aalborg University（奥尔堡大学）；Peking University（北京大学）；HKUST (Guangzhou)（香港科技大学（广州））；University of Cologne（科隆大学）；SenseTime Research（商汤科技研究院）

Chenxi Liu, Qianxiong Xu, Hao Miao, Sun Yang, Lingzheng Zhang, Cheng Long, Ziyue Li, Rui Zhao

备注：: Accepted as an Oral Presentation at AAAI 2025 (Main Technical Track)

阅读 PDF

arXiv 页面 ↗HTML 原文 ↗

AI总结：

TimeCMA通过跨模态对齐融合时间序列与文本提示，提取解纠缠且鲁棒的嵌入，在八个真实数据集上超越现有方法，实现高效多变量时间序列预测。

摘要：

多变量时间序列预测（MTSF）旨在学习变量之间的时间动态以预测未来的时间序列。现有的统计和基于深度学习的方法受限于可学习参数有限和训练数据规模较小。最近，将时间序列与文本提示相结合的大语言模型（LLMs）在MTSF中取得了令人鼓舞的性能。然而，我们发现当前基于LLM的解决方案在学习解纠缠嵌入方面存在不足。我们引入了TimeCMA，一个直观而有效的框架，通过跨模态对齐实现MTSF。具体而言，我们提出了一个双模态编码，包含两个分支：时间序列编码分支提取解纠缠但较弱的时间序列嵌入，而LLM赋能的编码分支将相同的时间序列与文本包装为提示，以获得纠缠但鲁棒的提示嵌入。因此，这种跨模态对齐基于时间序列和提示模态的相似性，从提示嵌入中检索出既解纠缠又鲁棒的时间序列嵌入，即“两全其美”。另一个关键设计是，为了减少时间序列及其长文本提示的计算成本，我们设计了一个有效的提示，鼓励最关键的时序信息封装在最后一个令牌中：只有最后一个令牌传递给下游预测。我们进一步存储最后一个令牌的嵌入以加速推理速度。在八个真实数据集上的大量实验表明，TimeCMA优于现有最先进方法。

英文摘要：

Multivariate time series forecasting (MTSF) aims to learn temporal dynamics among variables to forecast future time series. Existing statistical and deep learning-based methods suffer from limited learnable parameters and small-scale training data. Recently, large language models (LLMs) combining time series with textual prompts have achieved promising performance in MTSF. However, we discovered that current LLM-based solutions fall short in learning disentangled embeddings. We introduce TimeCMA, an intuitive yet effective framework for MTSF via cross-modality alignment. Specifically, we present a dual-modality encoding with two branches: the time series encoding branch extracts disentangled yet weak time series embeddings, and the LLM-empowered encoding branch wraps the same time series with text as prompts to obtain entangled yet robust prompt embeddings. As a result, such a cross-modality alignment retrieves both disentangled and robust time series embeddings, "the best of two worlds", from the prompt embeddings based on time series and prompt modality similarities. As another key design, to reduce the computational costs from time series with their length textual prompts, we design an effective prompt to encourage the most essential temporal information to be encapsulated in the last token: only the last token is passed to downstream prediction. We further store the last token embeddings to accelerate inference speed. Extensive experiments on eight real datasets demonstrate that TimeCMA outperforms state-of-the-arts.

机构、总结、中文摘要均由AI分析而来，可能存在错误，请以论文正文为准。