EvoFed: Leveraging Evolutionary Strategies for Communication-Efficient Federated Learning

Mohammad Mahdi Rahimi, Hasnain Irshad Bhatti, Younghyun Park, Humaira Kousar, Jaekyun Moon

Advances in Neural Information Processing Systems 36 (NeurIPS 2023) · 2023 · DOI: 10.52202/075280-2726

Abstract

Federated Learning (FL) is a decentralized machine learning paradigm that enables collaborative model training across dispersed nodes without forcing individual nodes to share data. Its broad adoption is hindered by the high communication costs of transmitting a large number of model parameters. EvoFed integrates Evolutionary Strategies with FL through fitness-based information sharing. Rather than exchanging updated model parameters, each node transmits distance-based similarity measures between its locally updated model and a synchronized population of noise-perturbed models. The server aggregates these fitness values to update the global model. The analysis establishes convergence, and experiments show performance comparable to FedAvg while drastically reducing communication requirements in practical settings.

Canonical citation

Rahimi, M. M., Bhatti, H. I., Park, Y., Kousar, H., & Moon, J. (2023). EvoFed: Leveraging Evolutionary Strategies for Communication-Efficient Federated Learning. Advances in Neural Information Processing Systems, 36.

@inproceedings{rahimi2023evofed,
  title = {EvoFed: Leveraging Evolutionary Strategies for Communication-Efficient Federated Learning},
  author = {Rahimi, Mohammad Mahdi and Bhatti, Hasnain Irshad and Park, Younghyun and Kousar, Humaira and Moon, Jaekyun},
  booktitle = {Advances in Neural Information Processing Systems},
  volume = {36},
  year = {2023},
  doi = {10.52202/075280-2726}
}

Archive note: this author-hosted PDF preserves the visible pages of the official version. Only embedded document metadata was normalized so title and author fields match the title page.