Mitigating multiple single-event upsets during deep neural network inference using fault-aware training