Figure 5
From: Inferring multi-target QSAR models with taxonomy-based multi-task learning
Performance on simulated data. Average mean squared error (MSE) while varying (a) the training set size N, (b) the number of tasks T, and (c) the number of attributes D. Varying a certain parameter, we kept the other parameters fixed to N=45, T=5, and D=14. The average MSE was calculated from the performance on the 10 randomly generated test data sets for each parameter setup. The upper graphs show results for high task similarity, the lower graphs for low task similarity.