… (DNNs) are very popular because of their high performance … of hardware approximation
techniques for DNN accelerators. … “Curable” approximation refers to approximation approaches …

… exhibit promising results when deployed in convolutional neural networks. By means of the
… design and use in hardware accelerators of CNNs. The approximations can be introduced to …

… design in which we additionally trade-off approximation with … loads in high performance
systems, DNN accelerators impose … , “CANN: Curable approximations for high-performance deep …

… In addition, we propose a filter-oriented approximation method to … [15] employs a curable
approximation in which the MAC's … in design of hardware accelerators of deep neural networks," …

… Recent Deep Neural Networks (DNNs) have brought … selected out of the neural networks
we consider in Section V. … ., “Cann: Curable approximations for high-performance deep …

… Abstract—Systolic array-based deep neural network (DNN) accelerators have recently …
., “CANN: Curable approximations for high-performance deep neural network accelerators,” …

… Specifically, we consider an inference accelerator similar to Google TPU [3], as depicted in
Figure 3, that employs a systolic MAC array and we replace the exact multipliers with LVRM. …

… Shafique, “Alwann: Automatic layer-wise approximation of deep neural network accelerators
… , “Cann: Curable approximations for high-performance deep neural network accelerators,” …

… for deep neural network accelerators. IEEE … of curable approximations is proposed in this
thesis. The concept enables the design of high-performance accelerators where approximation …

… systems, which is crucial in high-performance and low-power systems for … [16] proposed a
curable approximation Multiply-and-… of neural network algorithms and CNN accelerators, we …