ENCORE Compression: Exploiting Narrow-width Values for Quantized Deep Neural Networks | IEEE Conference Publication | IEEE Xplore