SYNTRA Enables Label-Free Cross-Domain Expansion of Embedding Models

A new arXiv paper introduces SYNTRA, a self-improving data synthesis method that generates domain-adapted training data for embedding models using only unlabeled corpora and a few query-document pairs. Validated on 24 datasets, it extends embeddings across 15 domains without manual labeling.

2026-10-14 ~ 2026-10-14 · 2 related posts

• SYNTRA: self-improving data synthesis scales embedding models across 15 domains — _reachsumit · 2026-10-14

1 near-duplicate retellings: _reachsumit