| Title | Time | Room | Teacher |
|---|---|---|---|
| Edge & Federated Learning | 09.07.2026 09:00 - 16:00 (Thu) | online | Gunnar Bless |
| Edge & Federated Learning | 10.07.2026 09:00 - 16:00 (Fri) | online | Gunnar Bless |
The seminar provides an overview of decentralized machine learning at the edge (edge computing) and federated learning. It covers the differences between cloud, edge and federated learning with a focus on IoT, data protection and resource constraints. Practical methods for model optimization and deployment on edge devices are presented and practiced. The architecture, algorithms and data protection techniques of federated learning are explained.
Foundations & Edge Computing
- Introduction & Setup - understand distributed ML paradigms
- Edge vs Cloud vs Federated Learning landscape
- Use cases: IoT, mobile devices, privacy-sensitive applications
- Challenges: resource constraints, communication costs, data heterogeneity
- Lab Compare standard vs compressed model performance. Measure inference time on simulated edge devices
Model Optimization for Edge
- Model compression techniques
- TensorFlow Lite conversion pipeline
- Hardware-specific optimizations
- Lab Quantize a CNN for image classification. Benchmark performance vs accuracy trade-offs
Edge Inference Implementation
- Edge deployment strategies
- Batch processing vs real-time inference
- Resource monitoring and management
- Lab Deploy quantized model using TFLite. Implement basic caching strategies
Distributed Edge Computing
- Coordinate multiple edge devices
- Load balancing and task distribution
- Fault tolerance in edge networks
Federated Learning Fundamentals
- FL architecture
- Federated Averaging FedAvg) algorithm
- Client-server architecture
- Privacy preservation techniques
- Communication efficiency strategies Lab FedAvg from scratch
Advanced FL Techniques
- Data heterogeneity and secure aggregation
- Personalization strategies
- Differential privacy in FL
- Secure aggregation protocols
- Lab: non-IID CIFAR-10 splits across clients. FedProx
Production FL Systems
- Production-ready FL
- Pipeline FL system architecture
- Client selection strategies
- Monitoring and debugging FL systems
- Integration with MLOps
- Pipelines Audit logging and compliance
Real-World Case Study
- Practical work on the PC
- Discussion and discussion of questions and tasks
- Lecture
Any phase
- Programming skills, preferably in Python
- Basic knowledge of machine learning and neural networks
- General mathematical and technical understanding
Participants will receive detailed information about the software to be provided before the course.
none
We send all participants a short questionnaire before the training to clarify previous knowledge and expectations.
In the event of late cancellation (less than 3 weeks before the start of the course) or no-show, an apology signed by your supervisor will be requested.