Federated Learning (FL) has emerged as a groundbreaking distributed machine learning paradigm—ideal for systems characterized by limited and intermittent connectivity. As modern satellite constellations in LEO grow in complexity, there is a clear trend toward integrating AI directly on-board rather than transmitting massive raw datasets back to Earth.

However, the reality of space is challenging: high mobility, intermittent links (ISL/GSL), and restricted bandwidth pose significant hurdles for traditional FL architectures.
Objective

The goal of this thesis is to investigate and develop methods for the efficient exchange and compression of gradients over satellite-to-satellite and satellite-to-ground links. You will work on ensuring that AI models can converge reliably despite the volatile network conditions inherent in orbital environments.
Research Focus

The project offers the opportunity for deep dives into several cutting-edge areas:

•    Gradient Compression: Implement and evaluate techniques such as quantization and sparsification to minimize communication overhead.
• Constellation Dynamics: Analyze how different constellation designs and communication capabilities impact the convergence of the learning process.
• Algorithm Development: Develop or adapt FL protocols specifically designed to handle the intermittent connectivity of LEO satellites.
• Simulation & Analysis: Use Python (e.g., PyTorch) to simulate the learning process within a dynamic satellite network environment.

Your Profile

We are looking for motivated students who enjoy bridging the gap between theory and application:

•    Degree Program: Student in Electrical Engineering, Computer Engineering, Computer Science, or a related field.
• Interests: Strong interest in Machine Learning and Satellite Communications.
• Programming: Solid skills in Python (ideally experience with PyTorch or TensorFlow).
• Background: Basic understanding of optimization or signal processing.