TableTime: Reformulating Time Series Classification as Training-Free Table Understanding with Large Language Models

TL;DR

This work tackles multivariate time series classification (MTSC) with large language models (LLMs) by addressing bottlenecks in numeric-to-text alignment, temporal/channel information preservation, and retraining costs. It introduces TableTime, a training-free framework that reformulates MTSC as a table-understanding task: time series are converted to tabular form, serialized to text, and classified via LLM reasoning augmented with neighbor guidance, domain context, task decomposition, and multi-path ensembles. The approach is validated on ten UEA MTSC datasets, showing competitive performance and strong data-efficiency, especially in small-sample scenarios, while outperforming several existing LLM-based methods. The results suggest that table-based representations and structured prompting can unlock robust, training-free MTSC with practical implications for resource-constrained settings and broad applicability.

Abstract

Large language models (LLMs) have demonstrated their effectiveness in multivariate time series classification (MTSC). Effective adaptation of LLMs for MTSC necessitates informative data representations. Existing LLM-based methods directly encode embeddings for time series within the latent space of LLMs from scratch to align with semantic space of LLMs. Despite their effectiveness, we reveal that these methods conceal three inherent bottlenecks: (1) they struggle to encode temporal and channel-specific information in a lossless manner, both of which are critical components of multivariate time series; (2) it is much difficult to align the learned representation space with the semantic space of the LLMs; (3) they require task-specific retraining, which is both computationally expensive and labor-intensive. To bridge these gaps, we propose TableTime, which reformulates MTSC as a table understanding task. Specifically, TableTime introduces the following strategies: (1) convert multivariate time series into a tabular form, thus minimizing information loss to the greatest extent; (2) represent tabular time series in text format to achieve natural alignment with the semantic space of LLMs; (3) design a reasoning framework that integrates contextual text information, neighborhood assistance, multi-path inference and problem decomposition to enhance the reasoning ability of LLMs and realize zero-shot classification. Extensive experiments performed on 10 publicly representative datasets from UEA archive verify the superiorities of the TableTime.

Figures (9)

• Figure 1: Illustrating the core idea of TableTime: transforming time series into tabular representations for universal classification in a data-centric paradigm.
• Figure 2: Illustration of the TableTime, i.e., a paradigm for MTSC based on table understanding. (Left): TableTime implements classification through neighbor retrieval and context information modeling. (Right): Multi-path ensemble enhancement.
• Figure 3: Prompt template of TableTime. There are three components of input from top to bottom.
• Figure 4: (Left) : Experimental results of four different neighbor retrieval methods: DTW, ED, SED, MAN. (Right) : Classification accuracy results under different neighbor number settings.
• Figure 5: Critical difference diagram over the mean ranks of TableTime, baseline models.