This research introduces a novel approach to semantic communication that eliminates the need to retrain models when switching between communication tasks.

• Leverages Mixture-of-Experts (MoE) architecture in large language models for zero-shot learning
• Enables wireless systems to handle new tasks without additional training
• Offers significant reduction in computational resources for wireless communication systems
• Improves flexibility by removing dependency on task-specific embeddings

For engineering applications, this breakthrough allows wireless systems to be more adaptable and efficient, potentially transforming how communication protocols are implemented in resource-constrained environments.

Leveraging MoE-based Large Language Model for Zero-Shot Multi-Task Semantic Communication