This thesis builds upon the recently proposed Cooperative Multi-Task Semantic Communication (CMT-SemCom) framework, which enables efficient and task-oriented information exchange for multiple classification tasks. The goal of this project is to extend the framework to support regression-based tasks, enabling cooperative learning and communication for both discrete and continuous inference problems.

The work will involve designing and implementing neural architectures capable of jointly handling classification and regression within a unified semantic communication setup, exploring information-theoretic objectives, representation learning, and multi-task cooperation. 

Requirements:

• Background in Machine Learning and Deep Learning (PyTorch preferred)

• Interest in semantic communication and multi-task learning

• A basic understanding of information theory and communication systems is advantageous