Replication of the paper "Adaptive dropout for training deep neural networks" (with reference to the related thesis). I apologise to the authors for the name "adaptive-standout", but I didn't want to call it "adaptive-dropout" as that would imply it was the only adaptive dropout algorithm; and I couldn't just call it "standout" as that seemed too general (makes it sound like a social networking product). So this name incorporates bits of both and makes less coherent sense than either. Should be OK for a name though.