Conference paperReducing the Impact of Phase-Change Memory Conductance Drift on the Inference of large-scale Hardware Neural Networks