Graphs have recently become more popular for modeling real-world
well as theoretical problems, and several Machine Learning algorithms
r learning on them have developed. For graphs that change
er time, however, research is still at an early stage. This article proposes
framework DynHeatmaP that generates a constant-size
aph sequence representation of graphs whose structure changes
er time (structural-dynamic, SDG) in linear time. The framework
es a representation of additions and deletions of nodes and edges
vents) to perform dynamic graph pooling. Our algorithm translates
e dynamic changes of the nodes and edges to a heatmap
presentation of the current node activity, regarding additions and
letions in the neighborhood, respectively. The created activity
atmap is utilized as probabilities of the nodes to be sampled in the
bsequent pooling step. In addition, the neighborhood of selected
des has been sampled such that the ratio of additions and deletions
represented accordingly. The obtained downsampled graph
quence can then be used representatively as input to a GNN for
arning on the original stream. Focussing on the active regions
a graph, the fast pooling process can reduce the training time
gnificantly and addresses the problem of unbalanced datasets in
al-world applications modeled as sparse graphs.