Duality in Multi-View Restricted Kernel Machines
This work provides theoretical unification of existing kernel methods, which is incremental but useful for researchers in kernel-based machine learning.
The authors developed a unified primal-dual multi-view framework that combines existing restricted kernel machine methods for kernel principal component analysis in supervised and unsupervised settings, demonstrating full equivalence between primal and dual formulations through theoretical derivation and experimental validation on time series forecasting tasks.
We propose a unifying setting that combines existing restricted kernel machine methods into a single primal-dual multi-view framework for kernel principal component analysis in both supervised and unsupervised settings. We derive the primal and dual representations of the framework and relate different training and inference algorithms from a theoretical perspective. We show how to achieve full equivalence in primal and dual formulations by rescaling primal variables. Finally, we experimentally validate the equivalence and provide insight into the relationships between different methods on a number of time series data sets by recursively forecasting unseen test data and visualizing the learned features.