Administrative Information
| Title | Federated Learning - Advances and Open Challenges |
| Duration | 45 - 60 |
| Module | C |
| Lesson Type | Lecture |
| Focus | Technical - Future AI |
| Topic | Advances in ML models through an HC lens - A result Oriented Study |
Keywords
Federated Learning,Decentralised data,Scalability,Non-convex optimization,Bias and Fairness,
Learning Goals
- Identify and discuss the advances of Federated Learning
- Recognise Open challenges of federated learning and discuss the proposed solutions
Expected Preparation
Learning Events to be Completed Before
Obligatory for Students
- Introduction to machine learning and deep learning concepts given in previous lectures
Optional for Students
None.
References and background for students
Recommended for Teachers
None.
Lesson materials
Instructions for Teachers
The goal of this lecture is to teach students how machine learning models can be refined when the model has been deployed on a device. This lecture should cover some of the basic concepts in FL but focus on the open problems, advances, and challenges outlined below.
Outline
| Duration | Description | Concepts | Activity | Material |
|---|---|---|---|---|
| 5 min | Federated learning (FL) lifecycle & training | Lifecycle (problem identification, instrumentation, prototyping, training, evaluation, deployment), Training (selection, broadcast, computation, aggregation, model update) | Taught session and examples | Lecture materials |
| 10 min | Algorithmic & practical challenges | Fully Decentralized / Peer-to-Peer Distributed Learning, SGD and network topologies, compression and quantization methods, Blockchain implementation of central server for aggregation, Cross-Silo (FL), Split learning | Taught session and examples | Lecture materials |
| 5 min | Efficiency & Effectiveness | Indepentant & identically distributed data (IID Data), Strategies for Dealing with Non-IID Data, Optimization Algorithms for FL | Taught session and examples | Lecture materials |
| 10 min | Model security (privacy & model attack) | Actors, Threat Models, Privacy in Depth, Secure Computations, Trusted execution environments, Local/Distributed/Hybrid differential privacy, Verifiability, External Malicious Actors | Taught session and examples | Lecture materials |
| 5 min | Fairness & Bias | Bias in Training Data, Fairness Without Access to Sensitive Attributes, Improving model diversity, | Taught session and examples | Lecture materials |
| 5 min | Systematic challenges | Development and deployment challenges, code deployment, monitoring and debugging, System induced bias, Parameter tunning | Taught session and examples | Lecture materials |
| 5 min | Conclusion, questions and answers | Summary | Conclusions | Lecture materials |
Acknowledgements
The Human-Centered AI Masters programme was Co-Financed by the Connecting Europe Facility of the European Union Under Grant №CEF-TC-2020-1 Digital Skills 2020-EU-IA-0068.