Boosting Graph Embedding on a Single GPU

Graphs are ubiquitous, and they can model unique characteristics complex relations of real-life systems. Although using machine learning (ML) on graphs is promising, their raw representation not suitable for ML algorithms. Graph embedding represents each node a graph as d-dimensional vector which more tasks. However, the process expensive, CPU-based tools do scale to real-world graphs. In this work, we present GOSH, GPU-based tool large-scale with minimum hardware constraints. GOSH employs novel coarsening algorithm enhance impact updates minimize work embedding. It also incorporates decomposition schema that enables any arbitrarily large be embedded single GPU. As result, sets new state-of-the-art in link prediction both accuracy speed, delivers high-quality embeddings classification at fraction time compared state-of-the-art. For instance, it embed over 65 million vertices 1.8 billion edges less than 30 minutes