k-Maximum Inner Product Attention for Graph Transformers and the Expressive Power of GraphGPS
This research introduces k-MIP attention to overcome the quadratic memory limits of graph transformers, enabling efficient processing of massive graphs while...