WaSeCom

WaSeCom is a model-agnostic framework for improving the reliability and adaptability of wireless semantic communication systems under semantic-level and channel-level uncertainty.

The project combines large AI models with Wasserstein distributionally robust optimization to preserve semantic fidelity under adversarial perturbations, distribution shifts, and challenging wireless channel conditions.

Research Focus

• Robust semantic encoding and decoding under semantic noise, adversarial perturbations, and input distribution shifts.
• Robust channel encoding and decoding under AWGN, Rayleigh fading, interference, and other channel variations.
• Bi-level Wasserstein distributionally robust optimization for semantic and physical layers.
• Model-agnostic integration with transformer-based and multimodal AI architectures.

Methodology

WaSeCom uses large AI model encoders for compact semantic representations across text, image, audio, and video modalities. A bi-level optimization formulation separates semantic robustness from channel robustness, while dual reformulation and smooth worst-case objectives make training scalable with deep learning models.

Contributions

• A Wasserstein robust framework for wireless semantic communication.
• Robustness guarantees for semantic and channel uncertainty.
• Model-agnostic design across large AI architectures and modalities.
• Empirical validation for image and text semantic communication tasks.

Related publication: (Le et al., 2025)

Collaborators

This project includes researchers from the University of Sydney, Kyung Hee University, Virginia Tech, Nanyang Technological University, University of Houston, Princeton University, and partner institutions.