BERT-JEPA: Reorganizing CLS Embeddings for Language-Invariant Semantics
Taj Gillin, Adam Lalani, Kenneth Zhang, Marcel Mateos Salles · 2026
Read BERT-JEPA: Reorganizing CLS Embeddings for Language-Invariant Semantics by Taj Gillin, Adam Lalani, Kenneth Zhang, Marcel Mateos Salles (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-2601-00366.txt.
Description
Taj Gillin, Adam Lalani, Kenneth Zhang, and Marcel Mateos Salles introduce BERT-JEPA, a training method for BERT-style language models. It adds a predictive embedding objective, drawn from joint embedding predictive architectures, to the usual training. The aim is to fix a collapsed summary-token embedding space and reshape it so that meaning is separated from language. The authors report that this new structure improves results across benchmarks in several languages.
Catalog record
| CREATOR | Taj Gillin, Adam Lalani, Kenneth Zhang, Marcel Mateos Salles |
|---|---|
| YEAR | 2026 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 35 KB |
| VERSION | 1.0 |
| LICENSE | http://creativecommons.org/licenses/by/4.0/ |
| RIGHTS | CC BY |
| SOURCE | arXiv |
Source: https://arxiv.org/abs/2601-00366