Federated Learning: Algorithms, Privacy, and Robustness 

Introduction to Federated Learning

• Overview of Federated Learning:

  • General Concepts (Privacy, Communication Efficiency, Robustness)

  • Key Challenges

  • FedAvg: Federated Averaging Algorithm

  • Implementation in PyTorch: Setting up the framework for experimentation

Advanced Algorithms (Statistical Challenges)

• FedProx: Addressing Statistical Heterogeneity

  • Complexity Analysis and Performance Trade-offs

• Flower Framework:

  • Comparative Analysis of FedAvg vs FedProx

Advanced Algorithms (System Heterogeneity)

• Fjord and HeteroFL: Handling System Heterogeneity in Federated Learning

• Flower Framework:

  • Exploring System Heterogeneity in Federated Learning

Practical Exercises (TP)

• Questions and Discussions on Implementations

Robustness in Federated Learning

• Adversarial Attacks:

  • Backdoor Attack, Label Flipping Attack, Model Poisoning, Byzantine Attack

• Defenses:

  • Median, Trimmed Mean, Nearest Neighbor Mean (NNM)

  • Attack Analysis and Defense Effectiveness

Privacy in Federated Learning (Attacks)

• Data Protection Regulations: Understanding the Legal Framework

• Threat Models: Identifying Key Vulnerabilities

• Privacy Attacks:

  • Reconstruction Attacks (Model Inversion, Gradient Inversion, Sample Convex Hull)

  • Membership Inference Attack, Source Inference Attack

  • Implementation: Hands-on exploration of attacks

Privacy in Federated Learning (Defense)

• Defense Mechanisms:

  • Differential Privacy

  • Secure Aggregation

  • Homomorphic Encryption

Evaluation & Grading

• Exam: 40%

• Practical Exercises (TP): 40%

• MiniTest: 20%