Fig. 9
From: Semisupervised regression in latent structure networks on unknown manifolds
Boxplot showing the superiority of our procedure over a one-stage procedure. The manifold is taken to be \({\mathcal {M}}=\psi ([0,L])\) where \(\psi (t)=\frac{1}{\sqrt{2}}(cos(t),sin(t),cos(t),sin(t))\). A total of 100 Monte Carlo samples of a random dot product graph of size \(n=500\) are generated, and the logarithms of the squared errors of each of the predicted responses are plotted. The red box corresponds to the predicted response from the true regressors, the orange box corresponds to the predicted response obtained from our method and the green box corresponds to the predicted response obtained from the one-stage procedure. It is evident from the figure that our procedure outperforms the one-stage procedure