Back to papers

Active Reinforcement Learning for Data Preparation: Learn2Clean with Human-In-The-Loop

Summary: Introduces Learn2Clean: a human-in-the-loop active reinforcement learning method that incrementally explores and prunes the combinatorial space of data-cleaning and preprocessing pipelines by querying users to guide actions for a target ML task. Frames data preparation as AI-hard, leveraging human feedback to trade off evaluation cost and downstream model quality, contrasting with passive AutoML/bandit approaches. (summarized by gpt-5-mini on Feb 09 2026)

Paper ID

380

Venue

CIDR

Year

2020

Pagerank

4.1945683e-05

Overall Rank

11,549 | 19.66%

DOI

-