Senior Researcher - Edge AI Optimization/Hardware-Aware ML

About the role

Huawei Canada’s Software‑Hardware System Optimization Lab is seeking a senior researcher to lead edge‑AI and hardware‑aware machine‑learning optimization. The role focuses on power‑efficiency and performance improvements for consumer devices across AI, multimedia, graphics, and mobile gaming.

Key responsibilities

• Conduct research on hardware‑aware neural‑network optimization techniques such as quantization‑aware training, mixed precision, pruning, distillation, and neural‑architecture search.
• Develop latency‑ and energy‑aware training objectives and Pareto‑optimization methods for accuracy, compute and memory trade‑offs.
• Prototype and evaluate efficient inference pipelines under device constraints (thermal limits, memory bandwidth, intermittent connectivity).
• Collaborate on compilers and runtimes (TVM, MLIR, XLA, TensorRT, ONNX Runtime, TFLite, ExecuTorch) to improve operator coverage and performance.
• Profile and optimise models with real device traces, addressing cache misses, kernel launch overhead, and CPU‑NPU hand‑off.
• Mentor junior researchers, review experimental designs and ensure measurement rigor and reproducibility.

Required profile

• PhD or equivalent research experience in Machine Learning, Computer Science, Electrical/Computer Engineering or a related field.
• 2+ years of research or industry experience with demonstrated impact in model compression, efficient architectures or ML systems/compilers.
• Strong publication record at top venues (NeurIPS, ICML, ICLR, MLSys, ASPLOS, ISCA, MICRO) and/or patents in ML efficiency.

Required skills

• Programming: Python, C/C++.
• Deep‑learning frameworks: PyTorch, TensorFlow, JAX.
• Deployment toolchains: ONNX, TFLite, TensorRT, TVM, MLIR‑based stacks.
• Performance profiling: latency measurement, memory profiling, kernel‑level bottleneck analysis.
• Hardware‑aware optimisation: quantization, pruning, sparsity, compiler graph rewriting, operator lowering, kernel autotuning.