Using Approximate DRAM for Enabling Energy-Efficient, High-Performance Deep Neural Network Inference | Publicación