Abstract
Principal component analysis (PCA) is a fundamental technique fordimensionality reduction and denoising; however, its application tothree-dimensional data with arbitrary orientations -- common in structuralbiology -- presents significant challenges. A naive approach requiresaugmenting the dataset with many rotated copies of each sample, incurringprohibitive computational costs. In this paper, we extend PCA to 3D volumetricdatasets with unknown orientations by developing an efficient and principledframework for SO(3)-invariant PCA that implicitly accounts for all rotationswithout explicit data augmentation. By exploiting underlying algebraicstructure, we demonstrate that the computation involves only the square root ofthe total number of covariance entries, resulting in a substantial reduction incomplexity. We validate the method on real-world molecular datasets,demonstrating its effectiveness and opening up new possibilities forlarge-scale, high-dimensional reconstruction problems.