σ-PCA: a building block for identifiable representations

σ-PCA is a special ICA method for eliminating the rotational indeterminacy from linear PCA — in other words, it can disentangle orthogonal components that share the same variance.