Exploring the intrinsic interconnections between the knowledge encoded in PRe-trained
Deep Neural Networks (PR-DNNs) of heterogeneous tasks sheds light on their mutual
transferability, and consequently enables knowledge transfer from one task to another so as
to reduce the training effort of the latter. In this paper, we propose the DEeP Attribution
gRAph (DEPARA) to investigate the transferability of knowledge learned from PR-DNNs. In
DEPARA, nodes correspond to the inputs and are represented by their vectorized attribution …

4 Setting TFi e→ tj to be the order of si; the nodes (DEPARA-V) or edges (DEPARA-E) by a
considerable margin, which again verifies the essentiality of both the nodes and the edges
for task transferability. These results are consistent with that of using taskonomy data as the
probe data. Except for these findings, another interesting observation is that DEPARA-V
outperforms DEPARA-E on COCO, but behaving worse on Indoor Scene. It indicates that for
different probe data, the relative importance of the nodes and the edges is changing for …