Summary of Test-time Training on Graphs with Large Language Models (llms), by Jiaxin Zhang et al.
Test-Time Training on Graphs with Large Language Models (LLMs)
by Jiaxin Zhang, Yiqi Wang, Xihong Yang, Siwei Wang, Yu Feng, Yu Shi, Ruicaho Ren, En Zhu, Xinwang Liu
First submitted to arxiv on: 21 Apr 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 A novel Test-Time Training pipeline, LLMTTT, is proposed to enhance test-time training on graphs with Large Language Models (LLMs) as annotators. This approach addresses the distribution shift between training and test data, which challenges Graph Neural Networks (GNNs). The LLMTTT pipeline introduces a hybrid active node selection strategy considering diversity, representativeness, and prediction signals from pre-trained models. A two-stage training strategy is designed to tailor the test-time model with limited and noisy labels. Experimental results show significant performance improvement compared to existing Out-of-Distribution generalization methods. |
| Low | GrooveSquid.com (original content) | Low Difficulty Summary GNNs have been successful in multimedia fields, but distribution shifts between training and test data can reduce their effectiveness. To overcome this, Test-Time Training (TTT) has been proposed. Traditional TTT methods require unsupervised training strategies, which are demanding. Inspired by Large Language Models’ ability to annotate Text-Attributed Graphs, a new approach enhances test-time training on graphs with LLMs as annotators. This novel pipeline, LLMTTT, uses a hybrid active node selection strategy to choose nodes for annotation. A two-stage training strategy is designed to adapt the model using limited and noisy labels. The method has been tested extensively and shown to improve performance significantly. |
Keywords
» Artificial intelligence » Generalization » Unsupervised
