Benchmarks on independent hold-out data showed that PARAS outperformed previous algorithms. Comparison of a wide range of both older and recent algorithms indicates that the volume of training data, not model choice, is the key determining factor of accuracy in these data regimes. 4/n
