Reproduction of multi-instance explanations and prototypes #34
Description
Dear authors,
Thanks for sharing the code for this nice paper!
I am particularly interested in the how you "obtain a global explanation of the class, which can shed light on how the identified structure for a given node is related to a prototypical structure unique".
As mentioned in your paper, it is done via, firstly, identifying a reference node for class c, v_c, and its associated important computation subgraph G_S(v_c), then aligning each of the identified computation subgraphs for all nodes in class c to the reference G_S(v_c). Utilizing the idea in the context of differentiable pooling [40].
I would like to reproduce your result in generating multi-instance / global explanation, would you also share that part of codes please? Many thanks!