Probabilistic Label Spreading: Efficient and Consistent Estimation of Soft Labels with Epistemic Uncertainty on Graphs
1 st Jonathan Klees, 2 nd Tobias Riedlinger, 3 rd Peter Stehr, 4 th Bennet Böddecker, 5 th Daniel Kondermann, 6 th Matthias Rottmann · 2026
Read Probabilistic Label Spreading: Efficient and Consistent Estimation of Soft Labels with Epistemic Uncertainty on Graphs by 1 st Jonathan Klees, 2 nd Tobias Riedlinger, 3 rd Peter Stehr, 4 th Bennet Böddecker, 5 th Daniel Kondermann, 6 th Matthias Rottmann (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-2602-04574.txt.
Description
Jonathan Klees and 5 co-authors present a cheaper way to estimate label uncertainty in a 2026 paper on probabilistic label spreading. Human annotations contain both inherent ambiguity and ignorance from limited knowledge, but collecting many labels per image is too costly at scale.
Their method spreads single annotations across a graph built from feature similarity, assuming nearby items share labels. They prove the resulting probability estimates stay consistent even as annotations per item approach zero. A scalable implementation follows. On common image datasets, the approach sharply reduced the annotation budget needed for a target label quality. It also set a new best result on a data-centric classification benchmark.
Catalog record
| CREATOR | 1 st Jonathan Klees, 2 nd Tobias Riedlinger, 3 rd Peter Stehr, 4 th Bennet Böddecker, 5 th Daniel Kondermann, 6 th Matthias Rottmann |
|---|---|
| YEAR | 2026 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 143 KB |
| VERSION | 1.0 |
| LICENSE | http://creativecommons.org/licenses/by/4.0/ |
| RIGHTS | CC BY |
| SOURCE | arXiv |
Source: https://arxiv.org/abs/2602-04574