Shelley Tong

Uncontrolled systematic errors have long barred neural representations from rigorous scientific inference—now, coupling contrastive learning with downstream nuisance parameterization forces latent manifolds to covary predictably under continuous physical distortions.