Digital Twin-Assisted Explainable AI for Robust Beam Prediction in mmWave MIMO Systems

Digital twins can reduce real-world data collection burdens: The paper demonstrates that synthetic channel data generated from digital twin simulations can effectively train AI models for wireless systems, significantly reducing the expensive and time-consuming process of collecting real-world data while maintaining performance quality through transfer learning refinement.

Explainability tools help identify what matters most: By using SHAP (Shapley Additive Explanations), the system can rank which input features are most important for beam prediction decisions, allowing engineers to understand why the AI makes specific choices and potentially reduce unnecessary data collection by focusing on key spatial directions.

Detecting when AI is uncertain improves reliability: The framework incorporates the Deep k-nearest neighbors algorithm to provide a "credibility metric" that flags when the AI encounters unfamiliar situations (out-of-distribution inputs), helping prevent overconfident predictions in novel real-world conditions and supporting safer deployment.

Robustness against adversarial attacks is essential for 6G infrastructure: The paper addresses susceptibility to adversarial attacks as a key challenge for deep learning in wireless systems, emphasizing that trustworthy AI for critical infrastructure like next-generation networks requires built-in defenses against malicious inputs.

Combining multiple safeguards creates more trustworthy systems: Rather than relying on a single approach, the paper integrates digital twins, explainability, transfer learning, and uncertainty detection—demonstrating that responsible AI implementation benefits from layered, complementary safeguards.