Duplo: Lifting Redundant Memory Accesses of Deep Neural Networks for GPU Tensor Cores | IEEE Conference Publication | IEEE Xplore