Machine learning non-Hermitian topological phases

Non-Hermitian topological phases have gained widespread interest due to their unconventional properties, which no Hermitian counterparts. In this work, we propose use machine learning identify and predict non-Hermitian phases, based on winding number. We consider two examples---non-Hermitian Su-Schrieffer-Heeger model its generalized version in one dimension nodal line semimetal three dimensions---to demonstrate the of neural networks accurately characterize phases. show that for one-dimensional model, a fully connected network gives an accuracy greater than 99.9% is robust introduction disorder. For three-dimensional find convolutional predicts different