Boundary-mask Jaccard analysis

methods / Theoretical / Analytical / Geometric analysis / Boundary-mask Jaccard analysis

Techniqueadvanced

Measures decision-boundary reconfiguration under a model perturbation (e.g. quantization) by rasterizing the boundary on a regular grid over prescribed 2D affine slices of input space and computing the Jaccard distance between the resulting boundary masks before and after the perturbation, alongside multiclass-junction topology where three or more class regions meet.

Used in (1 observation)

Quantization reconfigures decision-boundary geometry independent of accuracymeasured

structure: Decision boundary (as a codimension-1 hypersurface) · models: Small reduced-residual CNN classifier (CIFAR-10, full split, 3 seeds) · paper: Boundary-Aware Quantization: Finite-Scale Decision Geometry of Neural Classifiers