SYNTRA requires only an unlabelled corpus and a few relevant query-document pairs.
Self-improving data synthesis adapts embedding models to new tasks with minimal human input
SYNTRA iteratively refines shared synthesis instructions using feedback on trial outputs, enabling effective training data generation across diverse retrieval domains.
Academic
Zucan Lyu · Rui Tang · Jianlyu Chen · Peisheng Jin · Zheng Liu · Defu Lian
University of Science and Technology of China · State Key Laboratory of Cognitive Intelligence
Research Digest··3 min read
Thread:Agentic Data Synthesis
The authors propose SYNTRA, a method that generates task-specific training data for embedding models by iteratively improving synthesis instructions using feedback on trial outputs.
Why this paper
From University of Science and Technology of China and State Key Laboratory of Cognitive Intelligence · Part of Agentic Data Synthesis, now 4 papers
In one line
Self-improving data synthesis generates training data that adapts embedding models to new retrieval tasks across domains.
What we could check
- ·No code link found
- ·No weights link found
- ·No dataset link found
- ·No compute details found
- ✓Limitations stated by the authors (2 noted)
- ✓Reports numbers on named benchmarks (6 benchmarks)
Observed from the paper text and links we have. Absence here means we did not find it, not that it does not exist.
§