Communication-Efficient Federated Learning Systems

In distributed computing settings, data privacy is of paramount importance (e.g., mobile or medical devices). Federated Learning (FL) empowered the state-of-the-art deep neural network model to provide AI solutions to clients while keeping the client data private in distributed computing settings. Communication between different devices is one of the main bottlenecks of FL model because of the heterogeneity of computational resources and network conditions of different client devices. Considering the diversity and prevalence of AMD’s hardware ecosystem, we would like to evaluate and explore the generalizability and impact of various optimization methods (e.g., compression and quantization techniques). These optimization methods will focus on levigating communication bottlenecks in FL systems through improved computational intensity and memory usage. This study will provide AMD with an FL optimization method that AMD can apply to client FL systems solutions with AMD hardware.

Faculty Supervisor:

Nandita Vijaykumar

Student:

Partner:

AMD Canada

Discipline:

Computer science

Sector:

Manufacturing; Professional, scientific and technical services

University:

University of Toronto

Program: