# When Decentralized Optimization Meets Federated Learning

**Authors:** Hongchang Gao, My T. Thai, Jie Wu  
**arXiv:** [2306.02570](https://arxiv.org/abs/2306.02570)  
**Date:** 2023-06-05  
**Summary page:** https://www.summarizepaper.com/en/paper/2306.02570v1/

## Key points

- The article explores the challenges and opportunities of renovating decentralized optimization for federated learning
- Federated learning is a novel approach to extracting knowledge from distributed data that preserves privacy and reduces communication costs
- Most existing federated learning approaches are centralized, making them vulnerable to single-point failures
- The authors investigate the potential of a decentralized communication topology for federated learning to address this issue
- They systematically discuss the challenges and opportunities of this strategy from three perspectives: model, data, and communication
- Decentralizing federated learning can help improve its security while also allowing it to scale more efficiently with larger datasets
- This could potentially open up new possibilities for using federated learning in a variety of different settings where privacy is paramount or large amounts of data need to be processed quickly and securely.

## Summary

In their article titled "When Decentralized Optimization Meets Federated Learning," authors Hongchang Gao, My T. Thai, and Jie Wu explore the challenges and opportunities of renovating decentralized optimization for federated learning. Federated learning is a novel approach to extracting knowledge from distributed data that has gained popularity due to its ability to preserve privacy and reduce communication costs. However, most existing federated learning approaches are centralized, making them vulnerable to single-point failures. To address this issue, the authors investigate the potential of a decentralized communication topology for federated learning. They systematically discuss the challenges and opportunities of this strategy from three perspectives: model, data, and communication. By doing so, they aim to deepen our understanding of decentralized federated learning and provide insights into how it can be effectively implemented in various data analysis applications. Overall, their work highlights the importance of exploring alternative strategies for federated learning that can enhance its robustness and scalability in real-world scenarios. Through their research findings they demonstrate how decentralizing federated learning can help improve its security while also allowing it to scale more efficiently with larger datasets. This could potentially open up new possibilities for using federated learning in a variety of different settings where privacy is paramount or large amounts of data need to be processed quickly and securely.