Orthogonality-Explainability

A computer vision experiment measuring whether more orthogonal convolutional channels correspond to more separable visual concepts, using channel scores and labels for attribution.