ENSEMBLE A · AXES
μ [—] · Σ [—]
A toy linear denoising analogue: inspect what matching mean and covariance can explain before invoking memorization. This does not reproduce a diffusion model or the cited paper.
μ [—] · Σ [—]
μ [—] · Σ [—]
The matched ensembles returned identical `[0.600, −0.350]` outputs. Changing only covariance B to `diag(4, 0.25)` moved its output to `[0.960, −0.140]`, creating a measured gap of `0.417`.
Computation complete · runtime clean
Toy linear analogue; not a diffusion-model or paper reproduction.