On the Construction and Implications of Low-Loss Valleys in LoRA-based Bayesian Inference

Daniel Dold, Emanuel Sommer, Julius Kobialka, Oliver Dürr, David Rügamer · 2026

Read On the Construction and Implications of Low-Loss Valleys in LoRA-based Bayesian Inference by Daniel Dold, Emanuel Sommer, Julius Kobialka, Oliver Dürr, David Rügamer (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-2605-29580.txt.

Description

Daniel Dold and four colleagues ask whether low-loss valleys link separate optima in low-rank adaptation space, where discrete ensembles help little. They introduce LoRA-Curve, a Bézier curve in segmented free and anchored forms, the latter joining separately tuned optima. They prove continuity and Lipschitz regularity of loss along the curve. Tests show straight-line blends hit loss barriers, while anchored curves join optima through low-loss valleys. With flat-minima tweaks plus a divergence regularizer, the method raises predictive mutual information without losing performance.

Catalog record

CREATORDaniel Dold, Emanuel Sommer, Julius Kobialka, Oliver Dürr, David Rügamer
YEAR2026
COLLECTIONPAPERS
FORMATTXT
FILE SIZE63949
LICENSEhttp://creativecommons.org/licenses/by/4.0/
RIGHTSCC BY
SOURCEarXiv

Source: https://arxiv.org/abs/2605.29580

Browse the full library · About the Tower of Babel