Polynormer introduces a polynomial-expressive GT model with linear complexity to balance expressivity and scalability. It learns high-degree polynomials controlled by attention scores, achieving superior performance on multiple datasets. The architecture includes local and global equivariant attention models for learning node representations efficiently. Polynormer demonstrates the efficacy of linear local-to-global attention scheme in capturing critical global structures on graphs.
לשפה אחרת
מתוכן המקור
arxiv.org
תובנות מפתח מזוקקות מ:
by Chenhui Deng... ב- arxiv.org 03-05-2024
https://arxiv.org/pdf/2403.01232.pdfשאלות מעמיקות