Summary of Spatial-temporal Large Language Model For Traffic Prediction, by Chenxi Liu et al.
Spatial-Temporal Large Language Model for Traffic Prediction
by Chenxi Liu, Sun Yang, Qianxiong Xu, Zhishuai Li, Cheng Long, Ziyue Li, Rui Zhao
First submitted to arxiv on: 18 Jan 2024
Categories
- Main: Machine Learning (cs.LG)
- Secondary: Computation and Language (cs.CL)
GrooveSquid.com Paper Summaries
GrooveSquid.com’s goal is to make artificial intelligence research accessible by summarizing AI papers in simpler terms. Each summary below covers the same AI paper, written at different levels of difficulty. The medium difficulty and low difficulty versions are original summaries written by GrooveSquid.com, while the high difficulty version is the paper’s original abstract. Feel free to learn from the version that suits you best!
| Summary difficulty | Written by | Summary |
|---|---|---|
| High | Paper authors | High Difficulty Summary Read the original abstract here |
| Medium | GrooveSquid.com (original content) | Medium Difficulty Summary The proposed Spatial-Temporal Large Language Model (ST-LLM) for traffic prediction leverages the capabilities of large language models in time series analysis to outperform existing complex neural network structures. By defining timesteps at each location as tokens and designing a spatial-temporal embedding, the ST-LLM learns spatial location and global temporal patterns. A fusion convolution integrates these embeddings, while a partially frozen attention strategy adapts the LLM for capturing global dependencies. The model demonstrates robust performance in both few-shot and zero-shot prediction scenarios on real traffic datasets. |
| Low | GrooveSquid.com (original content) | Low Difficulty Summary Traffic prediction is used to forecast future traffic conditions at specific locations. This helps with intelligent transportation systems. Researchers have been trying to improve traffic prediction models, but existing ones haven’t seen much progress. Recently, big language models have shown great skill in analyzing time series data. This paper proposes a new model called Spatial-Temporal Large Language Model (ST-LLM) that uses this skill to predict traffic. The ST-LLM is different because it’s not just about adding more parameters or training for longer. It uses a special way of combining information from different locations and time points. |
Keywords
* Artificial intelligence * Attention * Embedding * Few shot * Large language model * Neural network * Time series * Zero shot
