Preprint
Aug 2026
Prototype-Rectified Iterative Self-supervised Manifold Denoising under Severe Acoustic Shift
PRISM (Prototype-Rectified Iterative Self-supervised Manifold Denoising), a training-free, source-free TTA framework grounded in the Affine Noise Hypothesis, is addressed, making it substantially faster than gradient-based TTA while requiring no additional training.
A. Shukla, R. Thakur, Aryan Das et al.
· 0 citations