Summary of Prompt-based Spatio-temporal Graph Transfer Learning, by Junfeng Hu et al.
Prompt-Based Spatio-Temporal Graph Transfer Learning
by Junfeng Hu, Xu Liu, Zhencheng Fan, Yifang Yin, Shili Xiang, Savitha Ramasamy, Roger Zimmermann
First submitted to arxiv on: 21 May 2024
Categories
- Main: Machine Learning (cs.LG)
- Secondary: Artificial Intelligence (cs.AI)
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 Spatio-temporal graph neural networks have shown promise in addressing complex dependencies for urban computing tasks like forecasting and kriging. However, their performance is limited by the need for extensive training data specific to a task, hindering adaptability across different domains with varying demands. To address this, we propose Spatio-Temporal Graph Prompting (STGP), a unified framework that enables cross-task generalization in spatio-temporal graph transfer learning. STGP unifies diverse tasks into a single template and introduces a task-agnostic network architecture aligned with this template, allowing for shared dependency capture. Learnable prompts are employed in a two-stage prompting pipeline to capture domain knowledge and task-specific properties, facilitating adaptation across domains. Our experiments demonstrate that STGP outperforms state-of-the-art baselines in forecasting, kriging, and extrapolation tasks, achieving up to 10.7% improvement. |
| Low | GrooveSquid.com (original content) | Low Difficulty Summary This paper is about using special kinds of neural networks for predicting things like weather or traffic patterns in cities. These networks are good at understanding complex relationships between different parts of the city. However, they need a lot of training data specific to each task, which makes it hard to use them for new tasks or in different cities. The researchers propose a new way to adapt these networks to different tasks and domains using something called “prompting”. This allows the network to learn from less data and be more flexible. They tested this approach on several tasks and showed that it works better than other methods, with improvements of up to 10.7%. |
Keywords
» Artificial intelligence » Generalization » Prompting » Transfer learning
