Skip to main content
. 2020 Apr 19;22(4):465. doi: 10.3390/e22040465

Figure 2.

Figure 2

The net4Lap Architecture. Given an input kNN graph, the process begins with a neural embedding process (stochastic gradient descent with negative sampling) yields a harmonic version that feeds the Laplacian regularizer step. The result is a denser graph suitable either for ranking or for obtaining an improved kNN graph which in turns feeds stochastic gradient descend for re-ranking.