Distributed Electromagnetic Neural Networks for Task-Oriented Semantic Communications

propose the Adaptive Probabilistic Matching Loss (APML), a fully differentiable