Jinrong Yang

Memory sharing, synchronization, and scheduling on heterogeneous hardware, with a focus on AI inference performance and resource efficiency.

1. 01

   Memory sharing on unified-memory hardware

   Studying data-sharing and synchronization costs on CPU–GPU unified-memory hardware, and designing memory-sharing libraries for different platforms.

2. 02

   CPU–GPU coordination for on-device inference

   Studying CPU–GPU synchronization and scheduling for on-device inference, balancing tail latency with CPU utilization.

3. 03

   OS structure for emerging accelerators

   Exploratory

   Exploring OS support for emerging accelerators such as neuromorphic chips, and the use of asymmetric multikernels for resource management.