An Evaluation of Backpropagation Interpretability for Graph Classification with Deep Learning A summary of the IEEE BigData 2020 research paper by Kenneth Teo Tian Shun, Eko Edita Limanta, and Arijit Khan [Graph Classification] Graph data are ubiquitous in many domains, such as social networks, knowledge graphs, biological networks, geo-spatial road networks, and internet-of-things. There are plenty of interest and applications in developing high-quality machine learning techniques for classifying graph objects, including cheminformatics (e.g., compounds that are active or inactive for some target) [1], [2] and bioinformatics (e.g., classifying proteins into different families) [3], malware detection and classification with call graphs [18]. There are two research directions for classification tasks on graphs: (1) Given a single large graph, infer labels of individual nodes (the node classification problem) [4], [5]; and (2) given a set of graphs with different struct