Finding non-linear relationships in data sets has become important in many fields of science. We compared the performances of CCA and three dense, kernelized CCA variants, KCCA, gradKCCA, (S)CCA-HSIC in 12 experiments with simulated data and two real-world scenarios. We concluded that the kernelized CCA variants are capable of retrieving complex non-linear relationships in simulated settings, and, in agreement with previous research, the methods can be applied in real-world scenarios.
