The Wristband Gaussian Loss: Deterministic, Composable Latents via a Sphere–Interval Decomposition
Mikhail Parakhin, André M. Carvalho, Patrick Haluptzok · 2026
Read The Wristband Gaussian Loss: Deterministic, Composable Latents via a Sphere–Interval Decomposition by Mikhail Parakhin, André M. Carvalho, Patrick Haluptzok (2026) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2605-08749.txt.
Description
Mikhail Parakhin, André Carvalho, and Patrick Haluptzok present a loss function that reshapes point embeddings into a perfect Gaussian without any random sampling. Each point splits into a direction on a sphere and a transformed radius, forming a wristband shape. They prove, with machine checked verification, that matching the uniform wristband target happens exactly when the source is a standard Gaussian, and they compute the matching energy two efficient ways.
Extra terms speed finite sample learning, and calibration turns the pieces into one standardized score. On tricky impostor data with correct margins but wrong dependence, the method leads at higher dimensions. Representation learning gains a deterministic route to clean latent spaces.
Catalog record
| CREATOR | Mikhail Parakhin, André M. Carvalho, Patrick Haluptzok |
|---|---|
| YEAR | 2026 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 72 KB |
| LICENSE | http://creativecommons.org/licenses/by/4.0/ |
| RIGHTS | CC BY |
| SOURCE | arXiv |
Source: https://arxiv.org/abs/2605.08749