SynQuE formalizes ranking synthetic datasets for real task performance with unannotated real data proxies, including LENS using LLM reasoning, showing correlations and accuracy gains like +8.1% on Text2SQL.
_eprint: https://onlinelibrary.wiley.com/doi/pdf/10.1890/08-2244.1
1 Pith paper cite this work. Polarity classification is still indexing.
1
Pith paper citing it
fields
cs.LG 1years
2025 1verdicts
CONDITIONAL 1representative citing papers
citing papers explorer
-
SynQuE: Estimating Synthetic Dataset Quality Without Annotations
SynQuE formalizes ranking synthetic datasets for real task performance with unannotated real data proxies, including LENS using LLM reasoning, showing correlations and accuracy gains like +8.1% on Text2SQL.