Predicting missing cell values in tabular data is a fundamental problem in data cleaning. While state-of-the-art reasoning models show great promise in predicting missing values in tables by reasoning holistically across rows and columns, they are costly to deploy at scale and tend to be overconfident, often generating hallucinated or false-positive predictions.
This paper observes that achieving high-precision missing-value prediction in tables requires a distinct combination of three capabilities:
- World knowledge
- Text-based reasoning
- Code-based reasoning
We systematically explore design choices for combining these capabilities and propose an Auto-Fill approach that post-trains three specialist small language models (SLMs), each optimized for one capability.
We develop a calibrated ensemble mechanism that dynamically selects the most confident specialist or abstains, ensuring high accuracy.
Extensive experiments on 11 benchmarks with 2200 real tables drawn from diverse domains show that Auto-Fill achieves superior accuracy compared to state-of-the-art reasoning models (e.g., o3-pro, Gemini 3 Pro, and DeepSeek R1), while operating at less than 1% of the cost of these frontier models.
Our results highlight the effectiveness of specialization and calibrated abstention in the important domain of tabular data. Auto-Fill is publicly available at GitHub.
Blogger's Review: The Auto-Fill method significantly enhances missing value prediction accuracy by integrating multiple reasoning capabilities. This specialized approach not only improves model performance but also reduces costs, showcasing the potential of effectively leveraging machine learning models in practical applications. This research provides new insights and directions for future data cleaning efforts.