BERT-JEPA: Reorganizing CLS Embeddings for Language-Invariant Semantics

Fuente: arXiv
Guardado en:
Detalles Bibliográficos
Autores principales: Gillin, Taj, Lalani, Adam, Zhang, Kenneth, Salles, Marcel Mateos
Formato: Preprint
Publicado: 2026
Materias:
Acceso en línea:
Etiquetas: Agregar Etiqueta
Sin Etiquetas, Sea el primero en etiquetar este registro!
_version_ 1866908744104804352
author Gillin, Taj
Lalani, Adam
Zhang, Kenneth
Salles, Marcel Mateos
author_facet Gillin, Taj
Lalani, Adam
Zhang, Kenneth
Salles, Marcel Mateos
contents Joint Embedding Predictive Architectures (JEPA) are a novel self supervised training technique that have shown recent promise across domains. We introduce BERT-JEPA (BEPA), a training paradigm that adds a JEPA training objective to BERT-style models, working to combat a collapsed [CLS] embedding space and turning it into a language-agnostic space. This new structure leads to increased performance across multilingual benchmarks.
format Preprint
id arxiv_https___arxiv_org_abs_2601_00366
institution arXiv
publishDate 2026
record_format arxiv
spellingShingle BERT-JEPA: Reorganizing CLS Embeddings for Language-Invariant Semantics
Gillin, Taj
Lalani, Adam
Zhang, Kenneth
Salles, Marcel Mateos
Computation and Language
Artificial Intelligence
Machine Learning
Joint Embedding Predictive Architectures (JEPA) are a novel self supervised training technique that have shown recent promise across domains. We introduce BERT-JEPA (BEPA), a training paradigm that adds a JEPA training objective to BERT-style models, working to combat a collapsed [CLS] embedding space and turning it into a language-agnostic space. This new structure leads to increased performance across multilingual benchmarks.
title BERT-JEPA: Reorganizing CLS Embeddings for Language-Invariant Semantics
topic Computation and Language
Artificial Intelligence
Machine Learning
url https://arxiv.org/abs/2601.00366