Tracking papers on "LLM for time series analysis", more details in Large Language Models for Time Series: A Survey.

[New🔥] Our survey paper is accepted to IJCAI 2024, survey track!

Left: Taxonomy of LLMs for time series analysis. If we outline typical LLM-driven NLP pipelines in five stages - input text, tokenization, embedding, LLM, output - then each category of our taxonomy targets one specific stage in this pipeline:

• Prompting (input stage) treats time series data as raw text and directly prompts LLMs with time series;
• Time Series Quantization (tokenization stage) discretizes time series as special tokens for LLMs to process;
• Aligning (embedding stage) designs time series encoder to align time series embeddings with language space;
• Vision as Bridge (LLM stage) connects time series with Vision-Lanuage Models (VLM) by employing visual representations as a bridge;
• Tool Integration (output stage) adopts language models to output tools to benefit time series analysis.

Right: Representative works for each category, sorted by their publication dates. The use of arrows indicates that later works build upon earlier studies. Dark(light)-colored boxes represent billion(million)-parameter models. Icons to the left of the text boxes represent the application domains of domain-specific models.

• Taxonomy
  • Prompting
  • Quantization
  • Aligning
  • Vision
  • Tool
• Datasets
• Citation

Left: VQ-VAE based quantization

Right: K-Means based quantization

Left: Aligning by similarity matching

Right: Aligning with large language models as backbones, where the output could be time series (e.g., forecasting) or text (e.g., EEG-to-text) depending on the downstream tasks.

If you find this useful, please cite our paper: "Large Language Models for Time Series: A Survey"

@article{zhang2024large,
  title={Large Language Models for Time Series: A Survey},
  author={Zhang, Xiyuan and Chowdhury, Ranak Roy and Gupta, Rajesh K and Shang, Jingbo},
  journal={arXiv preprint arXiv:2402.01801},
  year={2024}
}