This thesis focuses on extending Cooperative Multi-Task Semantic Communication (CMT-SemCom) into a federated learning (FL) paradigm to enable semantic multi-task learning across distributed space agents. The objective is to develop a unified framework where local models learn task-specific semantic representations and collaboratively update a shared global model without sharing raw data.

The project may also explore the semantic-aware clustering of tasks in the federated paradigm to examine the overall system performance.

Requirements:

• Background in machine learning or deep learning

• Experience with PyTorch

• Interest in semantic communication, multi-task learning, and federated systems

• Background in space engineering is a plus