Can Foundation Models Wrangle Your Data?
Avanika Narayan, Ines Chami, Laurel Orr, Simran Arora, Christopher Ré
TLDR
This paper demonstrates that large foundation models can effectively perform classical data cleaning and integration tasks without task-specific training, achieving state-of-the-art results.
Key contributions
- Reformulated five classical data cleaning and integration tasks as prompting tasks for foundation models.
- Showed that large foundation models generalize well and achieve state-of-the-art performance on these data tasks despite no specialized training.
- Identified challenges related to private and domain-specific data and highlighted opportunities to improve data management accessibility for non-experts.
Why it matters
This paper is important because it expands the application of foundation models beyond traditional language and image tasks into the realm of data management, suggesting that these models can simplify and improve data cleaning and integration processes. This has the potential to reduce the need for specialized expertise and costly task-specific model training, making data wrangling more accessible and efficient across diverse domains.
Original Abstract
Foundation Models (FMs) are models trained on large corpora of data that, at very large scale, can generalize to new tasks without any task-specific finetuning. As these models continue to grow in size, innovations continue to push the boundaries of what these models can do on language and image tasks. This paper aims to understand an underexplored area of FMs: classical data tasks like cleaning and integration. As a proof-of-concept, we cast five data cleaning and integration tasks as prompting tasks and evaluate the performance of FMs on these tasks. We find that large FMs generalize and achieve SoTA performance on data cleaning and integration tasks, even though they are not trained for these data tasks. We identify specific research challenges and opportunities that these models present, including challenges with private and domain specific data, and opportunities to make data management systems more accessible to non-experts. We make our code and experiments publicly available at: https://github.com/HazyResearch/fm_data_tasks.
📬 Weekly AI Paper Digest
Get the top 10 AI/ML arXiv papers from the week — summarized, scored, and delivered to your inbox every Monday.