ProtoSSL: Interpretable Prototype Learning from Unlabeled Time-Series Data
Steven Song, Sahil Sethi, Brett Beaulieu-Jones, Robert L. Grossman · 2026
Read ProtoSSL: Interpretable Prototype Learning from Unlabeled Time-Series Data by Steven Song, Sahil Sethi, Brett Beaulieu-Jones, Robert L. Grossman (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-2605-06943.txt.
Introduction
Steven Song and three co-authors introduce ProtoSSL, a way to pretrain interpretable prototype models without labels. Prototype networks explain predictions by comparing new cases with representative examples, which suits fields like medicine. Earlier versions needed labeled data, tying the learned examples to one task.
ProtoSSL first learns a bank of prototypes from unlabeled data, then assigns them to a downstream task with an efficient procedure. Across six heart-signal datasets, it beats supervised prototype methods when labels are scarce, and also after fine-tuning at full scale. Human judges preferred its explanations, and the framework also extends to audio classification.
Catalog record
| CREATOR | Steven Song, Sahil Sethi, Brett Beaulieu-Jones, Robert L. Grossman |
|---|---|
| YEAR | 2026 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 181 KB |
| LICENSE | http://creativecommons.org/licenses/by/4.0/ |
| RIGHTS | CC BY |
| SOURCE | arXiv |
Source: https://arxiv.org/abs/2605.06943