Robust Active Learning for Few-Shot Example Selection in Text-to-SQL

Name: Robust Active Learning for Few-Shot Example Selection in Text-to-SQL
Uploaded: 2026-10-03T11:57:25.746234+00:00
Duration: 762 s
Description: This work proposes a stratified greedy algorithm to select few-shot examples for text-to-SQL systems. It addresses challenges like noisy labels, the need for diverse examples across different topics, and imperfect mathematical models of the data space. The method maximizes information gain while ensuring each selected example comes from a unique semantic cluster.

This work proposes a stratified greedy algorithm to select few-shot examples for text-to-SQL systems. It addresses challenges like noisy labels, the need for diverse examples across different topics, and imperfect mathematical models of the data space. The method maximizes information gain while ensuring each selected example comes from a unique semantic cluster.

Open the full episode