• Multi-Perspective Feature Selection
Feature importance is usually more about a “local” conception than a “global”
one. To obtain a better representative feature subsets, the feature impact to
different low-embeddings or spectrums need to be considered [35]. Besides,
the view of instances is also indispensable since some features may only have
strong correlation with certain instances with respect to certain spectrums.
Therefore it is necessary to design a feature selection algorithm built upon
multi-perspective correlations. Our proposed algorithm selects features under
local context instead of global context, therefore it has a local view from both
the instance and feature perspectives, and measure their local correlations
with the global spectrums. Therefore, it provides a more informative feature
selection strategy.