Computing Platforms for Machine Learning
Machine learning applications require a tremendous amount of computing power. So, here we come. We are designing hardware-software co-optimized platforms to accelerate deep learning computations across a wide range of environments including IoT and datacenters.
- FPGA-based deep neural network acceleration
- Architecture design for reinforcement learning
- FPGA-GPU heterogeneous computing for neural network training
- GPU scheduling techniques
Spectre and Meltdown alarmed many computer architects and reminded the importance of interdisciplinary collaborations. We teamed up with computer security experts to tackle the processor vulnerability issues to provide more secure computing environments.
- Processor vulnerability detection and protection
- Error resilient robust systems architecture
As semiconductor technology scales and aims for higher efficiency, computer hardware platforms are exposed to many unreliability sources. We study the effects of such unreliability on system operation and provide effective set of tools to build reliable systems out of unreliable hardware components
- Application-level error resilience techniques
- Soft error mitigation techniques