Edge Models Optimization Research Engineer
Pubblicato il 17-08-2026 - Altro in Pisa
Position Summary:
Huawei's Pisa Research Institute is a core innovation hub in the field of automotive software and embedded systems. We are looking for a highly motivated researcher to join our AI platform team and contribute to the development of next-generation models for intelligent vehicles and edge computing systems. The successful candidate will focus on post‐training techniques, efficient Mixture‐of‐Experts (MoE) architectures, and deployment optimization on resource‐constrained platforms. Working closely with system software, compiler, and hardware teams, the candidate will help bridge cutting‐edge AI research and real‐world automotive applications. Key Responsibilities:
Investigate advanced post‐training approaches for large language models and multimodal foundation models, including supervised fine‐tuning, preference alignment, reinforcement learning‐based optimization, and knowledge distillation.
Explore efficient MoE architectures and sparse inference mechanisms that enable high‐quality intelligence under strict latency, memory, and power constraints.
Develop and evaluate model compression techniques such as quantization, pruning, and low‐rank adaptation, while also optimizing inference performance across heterogeneous computing platforms including CPUs, GPUs, NPUs, and automotive SoCs.
The role further involves benchmarking state‐of‐the‐art open‐source models, collaborating with software platform teams to improve deployment efficiency, and contributing to research publications, patents, and technical innovation initiatives.
Basic Qualifications:
Candidates should hold a master or PhD in Computer Science, Artificial Intelligence, Machine Learning, Computer Engineering, or a closely related field.
A strong understanding of transformer architectures, large language models, and modern deep learning techniques is required.
Experience with PyTorch and distributed training frameworks is expected, while previous research experience in model optimization, effic