aggregation readout layer

[[concept]]

Aggregation Readout Layer

The aggregation readout layer is a readout layer given by $y=\text{aggr}(x_{L}C)$

where $C\in {\mathbb{R}}^{d_L\times d}$ and $\text{aggr}:{\mathbb{R}}^{n\times d}\to {\mathbb{R}}^{1\times d}$ is node-level aggregation Typical aggregation functions are the mean, sum, max, min, median, etc

NOTE

• this is now independent of $n$ (my aggregation functions don’t change based on the size of my graph)
• permutation invariant since the aggregation functions are also permutation invariant
• this remains transferrable across graphs, unlike the fully connected readout layer

Mentions

TABLE
FROM [[]]
 
FLATTEN choice(contains(artist, this.file.link), 1, "") + choice(contains(author, this.file.link), 1, "") + choice(contains(director, this.file.link), 1, "") + choice(contains(source, this.file.link), 1, "") as direct_source
 
WHERE !direct_source