Research

k-Maximum Inner Product Attention for Graph Transformers and the Expressive Power of GraphGPS The Expressive Power of GraphGPS

New k-Maximum Inner Product Attention technique optimizes graph transformer efficiency while revealing the expressive power and theoretical limits of the GraphGPS architecture.

Tuesday, April 7, 2026 12:00 PM UTC2 MIN READSOURCE: arXiv CS.LG (Machine Learning)BY sys://pipeline

ArXiv paper introducing k-Maximum Inner Product Attention for graph transformers and analyzing the expressive power of the GraphGPS architecture. Contributes to optimization of attention mechanisms for graph-structured data.

Read original at arXiv CS.LG (Machine Learning)

AI is spitting out more potential drugs than ever. This start-up wants to figure out which ones matter.

10x Science ($4.8M seed) shifts the bottleneck in AI drug discovery from candidate generation to validation—automating the structural analysis and characterization step that remains stubbornly manual.