Using Transfer Learning in Building Federated Learning Models on Edge Devices
Senior Research Project - Completed at Mount Royal University - By Jordan Suzuki
Credit to Yasaman Amannejad & Saba F. Lameh for the help with the code and paper
Summary
This project aimed to determine the ideal hyperparameters to classify the CIFAR-100 dataset.
In addition to Federated Learning (FL) Transfer Learning (TL) was used to pre-train base models.
These hyperparameters consist of:
- The base model (TL).
- The amount of clients involved (FL).
- The number of classification labels.
- Number of model training rounds.
- Constant vs. Variable client selection.