Transformers are efficient hierarchical chemical graph learners
Pengmei, Zihan, Li, Zimu, Tien, Chih-chan, Kondor, Risi, Dinner, Aaron R.
–arXiv.org Artificial Intelligence
Transformers, adapted from natural language processing, are emerging as a leading approach for graph representation learning. Contemporary graph transformers often treat nodes or edges as separate tokens. This approach leads to computational challenges for even moderately-sized graphs due to the quadratic scaling of self-attention complexity with token count. In this paper, we introduce SubFormer, a graph transformer that operates on subgraphs that aggregate information by a message-passing mechanism. This approach reduces the number of tokens and enhances learning long-range interactions. We demonstrate SubFormer on benchmarks for predicting molecular properties from chemical structures and show that it is competitive with state-of-the-art graph transformers at a fraction of the computational cost, with training times on the order of minutes on a consumer-grade graphics card. We interpret the attention weights in terms of chemical structures. We show that SubFormer exhibits limited over-smoothing and avoids over-squashing, which is prevalent in traditional graph neural networks.
arXiv.org Artificial Intelligence
Oct-2-2023
- Country:
- North America > United States
- New York > New York County
- New York City (0.14)
- Massachusetts
- Suffolk County > Boston (0.04)
- Middlesex County > Cambridge (0.04)
- Illinois > Cook County
- Chicago (0.05)
- California > Los Angeles County
- Pasadena (0.04)
- New York > New York County
- Asia > China
- North America > United States
- Genre:
- Research Report > New Finding (0.46)
- Technology: