Data-free knowledge distillation

  • Designed a new loss function for the generator to produce samples with high authenticity, class diversity, and inter-sample diversity to facilitate data-free knowledge distillation with higher accuracy and more robustness over different hyperparameter settings

Collaborated with Jihong Park (Deakin University, Australia) and Shiqiang Wang (IBM T.J. Watson Research Center, USA)

Service placement in mobile edge cloud networks

  • Formulated the interference in terms of delay among services coexisting on the same infrastructure; Designed algorithms to minimize overall resource cost for multi-component service placement to achieve interference awareness and delay guarantee

  • Defined the equivalent bandwidth for the flexible resource allocation of an edge in a service; Formulated and resolved the joint optimization of small cell zooming and multi-component service placement with reduced resource fragments and guaranteed QoS

Communication-efficient federated learning

  • Automatically determined the degree the sparsity of gradient sparsification enabled federated learning with non-i.i.d. local datasets to minimize the overall training time

Collaborated with Shiqiang Wang (IBM T.J. Watson Research Center, USA) and Kin K Leung (Imperial College London, UK)

Capacity of distributed computing systems

  • Derived theoretical formulas for the capacity of distributed systems with multiple resource types, where the power of d choices is considered as the task scheduling strategy in the analysis

Collaborated with Kin K Leung (Imperial College London, UK) and Shiqiang Wang (IBM T.J. Watson Research Center, USA)

Resource allocation in converged fiber-wireless (FiWi) access networks

  • Focused on improving the physical resource utilization in Virtual Network Embedding (VNE) applications.Adjusted the QoS satisfaction of virtual networks to achieve best-effort service and increase the revenue of infrastructure provider
  • Proposed energy-saving VNE schemes for 5G enhanced FiWi access networks with scalable cells
  • Proposed algorithms for controlling the power states of network devices for energy-efficient network design