Researcher - AI Computing System

The Advanced Computing and Storage Lab, currently a part of the Vancouver Research Centre, aims to explore adaptive computing system architectures to address the challenges posed by flexible and variable application loads in the future. It assists in ensuring the stability and quality of training clusters, constructs dynamic cluster configuration strategy solvers, and establishes precision control systems to create stable and efficient computing power clusters. One of the lab's goals is to focus on key industry AI application scenarios such as large model training/inference, based on key technologies like low-precision training, multi-modal training, and reinforcement learning, responsible for bottleneck analysis and the design and development of optimization solutions, thereby improving training and inference performance as well as usability.

• Aiming at key industry AI application scenarios such as large model training and inference, this role focuses on advancing performance, efficiency, and usability of AI systems on the Ascend platform. The work involves low-precision training, multimodal optimization, reinforcement learning, and training resource optimization to address system bottlenecks and deliver next-generation AI capabilities.

• Responsible for design and development of optimization solutions for AI training and inference systems, with a focus on FP8 optimization, RL-driven training agents, multimodal reinforcement learning or next-generation multi-modal understanding & generation.

• Combine AI algorithm requirements with system-level architectural optimization in computing, I/O, scheduling, and precision control to improve performance.

• Build stable, efficient AI training clusters, leveraging dynamic cluster configuration and precision control to ensure scalability and reliability.

• Develop software frameworks, operator libraries, acceleration libraries, and system-level optimizations for NPU platforms to accelerate large-model AI training.

• Drive innovation in optimizing large-model training and inference with low-precision training, parallel strategy tuning, and reinforcement learning.

• Grasp the latest research progress and technological trends in AI computing cluster architecture design, training acceleration, and inference acceleration across academia and industry to strengthen the competitiveness of AI computing cluster systems.

The target annual compensation (based on 2080 hours per year) ranges from $78,000 to $150,000 depending on education, experience and demonstrated expertise.