Graph Isomorphism Network: A Learning-based Workflow for Converter Inverse Design Problem

To surpass the limitation of conventional artificial intelligence-enabled automatic design methodology, this work proposes a novel workflow to implement the emerging graph-learning-based method called Graph Isomorphism Networks (GIN) for power converter automated design. A radio-frequency filter design demo is provided to show the merits of bringing GIN into converter design, where similar tasks could take days if not weeks to achieve optimally by hand.