Abstract: One-shot neural architecture search (NAS ... To overcome the issue, we propose an orthogonal gradient learning (OGL) guided supernet training paradigm, where the novelty lies in the fact ...