Federated Learning: What is it?

Federated learning (FL) is a way to train AI models without anyone(seeing or touching your data), offering a way to unlock information to feed new AI applications.

But then what about Deep Learning ! and Distributed Deep Learning !. Right, at first, FL might seem like a new buzzword for distributed deep learning. But it is not.

The main difference is that in FL, the data is not shared. Instead, the model is shared and trained on the data that remains on the device.

Type of FL: Based on the Scope

• Cross Silio Federated Learning
• Cross Device Federated Learning
• Cross Organization Federated Learning

Type of FL:Based on Data Partitioning

• Horizontal Federated Learning
• Vertical Federated Learning
• Federated Transfer Learning

Federated Learning Architecture

• Centralized
• Decentralized

Federated Learning Challenges

• System Heterogeneity

Different devices have different hardware and software.

• Statistical Heterogeneity

Different devices have different data distributions.

• Communication Efficiency

The model needs to be shared and trained on the data that remains on the device.

• Privacy and Security

The data remains on the device and is not shared with the server.

Recent Advanced Research Topics in Federated Learning

• Security Concerns
• Adversarial Attacks
• Byzantine Robustness
• Privacy-Preserving Mechanisms
  • Differential Privacy
  • Homomorphic Encryption
• Novel Aggregation Techniques
• Communication Efficiency
• Model Compression
• Asynchronous FL
• Domain Specific FL
  • Edge Computing/IoT
  • Healthcare
  • Autonomous System
• Federated Reinforcement Learning (FRL)
• Multi-Modal FL
• Federated Transfer Learning