Zihan Zhang, Ph.D.

Building edge–cloud AI infrastructure at Lenovo AI Technology Centre (LATC), Edinburgh.

Researched resource-efficient, privacy-preserving distributed machine learning on edge devices.

• Built a distributed ML system for training language models and CNNs across 100+ edge devices; reduced end-to-end training time and improved model accuracy.
• Overlapped computation and communication using pipeline parallelism and asynchronous aggregation, reducing communication, on-device computation, and idle time; stabilised throughput under stragglers and intermittent connectivity.
• Collaborated with Rakuten Mobile Japan on US patent applications and peer-reviewed publications.

Co-founded a project exploring gaze-based authentication on everyday devices, without additional equipment.

• Co-authored the successful CyberASAP bid; our team secured £92k to develop a gaze-based authentication PoC and roadmap addressing GenAI spoofing threats.
• Built the authentication pipeline from capture and calibration through gaze estimation, embedding, and matching, with challenge–response and basic liveness checks.
• Collected 100+ participant sessions on phones and laptops and presented the PoC at CyberASAP Demo Day.

Applied machine learning to supply-chain optimisation at Huawei Noah's Ark Lab.

• Delivered a distributed ML-driven packing and routing system that improved average packing rate and reduced annual logistics costs by several million USD in production.
• Built a configurable constraint framework and learning-based solver as a tested algorithm library with defined APIs; the platform team containerised and deployed it on Huawei Cloud.
• Co-organised the EMO 2021 competition, providing a Docker image, automated evaluation and reproducible baseline; open-sourced the winning solution on GitHub.
• Mentored an intern and co-authored an ACM CIKM paper on ML for large-scale optimisation.