Adaptable Approximation Based on Bit Decomposition for Deep Neural Network Accelerators | IEEE Conference Publication | IEEE Xplore