Semantically Interpretable Activation Maps: what-where-how explanations within CNNs | IEEE Conference Publication | IEEE Xplore