Genetic CFL: Hyperparameter Optimization in Clustered Federated Learning.

Area of Science:

• Artificial Intelligence
• Machine Learning
• Distributed Systems

Background:

• Federated learning (FL) offers a distributed approach for deep learning but faces implementation challenges.
• Key limitations include technological constraints, communication overhead, non-IID data, and privacy concerns.
• Training on heterogeneous non-IID data significantly degrades model convergence and performance.

Purpose of the Study:

• To address inefficient client training and static hyperparameter use in existing FL algorithms.
• To propose a novel hybrid algorithm, Genetic CFL, for improved federated learning.
• To enhance individual cluster accuracy through integrated clustering and hyperparameter optimization.

Main Methods:

• Developed a hybrid algorithm, Genetic CFL, clustering edge devices by training hyperparameters.
• Employed genetic algorithms for cluster-wise parameter modification.
• Integrated density-based clustering with genetic hyperparameter optimization to boost accuracy.

Main Results:

• Genetic CFL demonstrated significant improvements in performance on benchmark datasets.
• Achieved 99.79% accuracy on the MNIST dataset and 76.88% on CIFAR-10.
• Effective performance was observed even with non-IID and ambiguous data within 10 training rounds.

Conclusions:

• The proposed Genetic CFL effectively overcomes limitations of traditional and clustered FL.
• The algorithm shows strong potential for practical implementation in real-world scenarios with non-IID data.
• This hybrid approach enhances model accuracy and convergence in federated learning environments.