Semantic Correlations Loss: Improving Model Interpretability for Multi-class Classification | IEEE Conference Publication | IEEE Xplore