DSpodFL

Decentralized Sporadic Federated Learning: A Unified Algorithmic Framework with Convergence Guarantees

• arXiv: Decentralized Sporadic Federated Learning: A Unified Algorithmic Framework with Convergence Guarantees

Instruction

• Agent.py: Implements an edge device in the network, which conducts local gradient descent and local aggregation with its neighbors.
• DSpodFL.py: Main code which handles running the DSpodFL methodology, where the network graph is generated and SGD and aggregation probabilities are assigned to nodes and edges, respectively.
• main.py: Different experiments under various setups that are reported in the paper are obtained by running this code. Simply change the specifications in the _init_() function and run this code.

Citation

@article{zehtabi2024decentralized,
  title={Decentralized Sporadic Federated Learning: A Unified Methodology with Generalized Convergence Guarantees},
  author={Zehtabi, Shahryar and Han, Dong-Jun and Parasnis, Rohit and Hosseinalipour, Seyyedali and Brinton, Christopher G},
  journal={International Conference on Learning Representations (ICLR)},
  year={2024}
}