Auto-Fill: Learning to Predict Missing Values Accurately with Specialist Language Models

2026-07-22

AI Summary

This work addresses the limitations of large language models in missing value imputation for tabular data—namely, insufficient accuracy, high inference costs, and susceptibility to hallucination—by proposing a capability-partitioned ensemble of small expert language models. Specifically, three specialized small models are post-trained to excel respectively in world knowledge, textual reasoning, and code-based reasoning. A calibrated ensembling strategy combined with a dynamic abstention mechanism enables the system to select the most suitable expert on a per-query basis or abstain when confidence is low. Evaluated across 11 benchmarks encompassing 2,200 real-world tables, the proposed approach significantly outperforms state-of-the-art models such as o3-pro and Gemini 3 Pro in imputation accuracy while reducing inference costs to less than 1% of theirs.

Abstract

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. In this paper, we observe that achieving high-precision missing-value prediction in tables requires a distinct combination of three capabilities: (1) world knowledge, (2) text-based reasoning, and (3) 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 either 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 a fraction (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 https://github.com/lyrain2001/auto-fill.

Problem

• missing value prediction
• tabular data
• data cleaning
• hallucination
• model overconfidence

Innovation

• specialist language models
• missing value imputation
• calibrated ensemble
• tabular data reasoning
• cost-efficient AI