Federated Collaborative Learning with Sparse Gradients for Heterogeneous Data on Resource-Constrained Devices
Author Information
Author(s): Li Mengmeng, He Xin, Chen Jinhua, Kotsiantis Sotiris
Primary Institution: Henan University, Kaifeng, China
Hypothesis
Can a sparse gradient collaborative federated learning model improve training efficiency and accuracy for heterogeneous data on resource-constrained devices?
Conclusion
The proposed algorithm significantly improves classification efficiency and addresses challenges posed by data heterogeneity in federated learning.
Supporting Evidence
- The proposed algorithm reduces communication overhead and improves training efficiency.
- Experiments show that the method enhances model accuracy while maintaining privacy.
- The framework allows for parallel training on resource-constrained devices.
Takeaway
This study shows how devices can work together to learn without sharing their data, making it faster and safer, especially when they have different types of data.
Methodology
The study designs a federated collaborative learning framework that combines sparse gradient strategies with adaptive aggregation weights to improve training efficiency and accuracy.
Potential Biases
Potential for imbalanced contributions from clients due to data heterogeneity.
Limitations
The method may expose sensitive user information and could be vulnerable to malicious attacks.
Digital Object Identifier (DOI)
Want to read the original?
Access the complete publication on the publisher's website