AI Infrastructure and Tools

Wolong: A Backend Optimizer for Deep Learning Computation and Open PAI

Chair: Jilong Xue

This talk presents Wolong, a backend optimization system for accelerating deep learning computation on graphics processing units (GPUs). It automatically applies operator batching and kernel fusion to avoid system overhead such as scheduling and kernel launch by analyzing the dataflow graph. This optimization is transparent to the AI model developer and, by integrating with TensorFlow, it can accelerate deep learning computation rates by an order of magnitude.

Open PAI: Open Source Initiative for AI Platform in China

Chair: Fan Yang

Open Platform for AI (Open PAI) is an open source platform for graphics processing unit (GPU) cluster management and resource scheduling. The platform’s design has been proven in Microsoft’s large-scale production environment.

Series: Microsoft Research Faculty Summit