Self-Improving Interference Management Based on Deep Learning With Uncertainty Quantification

This paper presents a groundbreaking self-improving interference management framework tailored for wireless communications, which integrates deep learning (DL) with uncertainty quantification (UQ) to enhance overall system performance. A key breakthrough of our framework is to mitigate the limitations inherent in data-driven interference management DL models, particularly in scenarios not adequately represented by the training dataset. To this end, we design a qualifying criterion that assesses the trustworthiness of DL model-based solutions by utilizing a UQ method. Guided by the criterion, the framework strategically alternates between DL model-based solutions and traditional algorithm-based solutions to maximize the credibility of the solutions in a cost-effective manner. Experimental results validate the effectiveness of our framework and demonstrate its superiority over traditional DL-based interference management, notably in scenarios underrepresented in the training dataset. This work marks a pioneering effort in harnessing self-improving DL for interference management, through the lens of UQ.