Model-based deep learning

Domain-inspired learning on graphs

Domain-inspired learning on graphs

Developed a hybrid framework to augment an iterative power allocation algorithm (WMMSE) with graph neural network (GNN) based learning modules for fast and efficient power allocation in SISO (Paper,Paper) and MIMO (Paper,Paper) wireless networks. Theorized and empirically validated stability bounds on the hybrid algorithm with respect to input perturbations (Paper).