Reproduce “Continual Learning of Domain Knowledge from Human Feedback in Text-to-SQL” on OpenResearch

Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.

Universal app for Apple silicon and Intel. Requires macOS 11 or later.

About this paper

Thomas Cook, Kelly Patel, Sivapriya Vellaichamy, Udari Madhushani Sehwag, Saba Rahimi, Zhen Zeng, Sumitra Ganesh

A framework from JPMorganChase enables large language models to continually learn domain knowledge from human feedback in text-to-SQL tasks. It leverages a multi-level memory and procedural reasoning to distill tacit human expertise, improving SQL generation accuracy and generalization to new queries.

Key insights

  • Procedural reasoning in the learning agent is crucial for effectively utilizing distilled knowledge and multi-level memory, enhancing both memory retention and generalization capabilities.
  • Distilling tacit domain knowledge from human feedback, when appropriately integrated via procedural reasoning, leads to improved semantic correctness in generated SQL queries.
  • A multi-granular memory structure, storing different types of knowledge, enables more robust and adaptive continual learning for text-to-SQL agents.

Summary by alphaXiv.

Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.