CFMI: Flow Matching for Missing Data Imputation

2025-06-10

AI Summary

To address the challenges of generality, modeling capacity, and scalability in missing value imputation for tabular and time-series data, this paper proposes Conditional Flow Matching Imputation (CFMI). CFMI is the first method to integrate conditional flow matching with a shared conditional encoder, yielding a differentiable, efficient, and high-dimensional scalable generative imputation framework that supports zero-shot time-series imputation. Technically, it unifies continuous normalizing flows and flow matching under a single conditional modeling paradigm, enabling joint optimization across data modalities. On 24 medium- and small-scale tabular benchmarks, CFMI matches or surpasses nine state-of-the-art baselines. For zero-shot time-series imputation, it achieves accuracy comparable to diffusion models while accelerating inference by several-fold. In high-dimensional settings, CFMI maintains robust performance, effectively alleviating the limitations of traditional multiple imputation—namely, inadequate modeling of complex dependencies and poor scalability.

Technology Category

• Machine Learning: Time-Series/Data Streams
• Computer Vision: Multi-modal Vision
• Reasoning under Uncertainty: Relational Probabilistic Models

Application Category

• User Modeling, Personalization and Recommendation: Explainable and interpretable methods for personalization
• Graph Algorithms and Modeling for the Web: Representation, reconstruction, and subgraph or motif discovery in Web-related graphs
• Search and Retrieval-Augmented AI: Retrieval-Augmented Generation (RAG) and multi-modal RAG

Abstract

We introduce conditional flow matching for imputation (CFMI), a new general-purpose method to impute missing data. The method combines continuous normalising flows, flow-matching, and shared conditional modelling to deal with intractabilities of traditional multiple imputation. Our comparison with nine classical and state-of-the-art imputation methods on 24 small to moderate-dimensional tabular data sets shows that CFMI matches or outperforms both traditional and modern techniques across a wide range of metrics. Applying the method to zero-shot imputation of time-series data, we find that it matches the accuracy of a related diffusion-based method while outperforming it in terms of computational efficiency. Overall, CFMI performs at least as well as traditional methods on lower-dimensional data while remaining scalable to high-dimensional settings, matching or exceeding the performance of other deep learning-based approaches, making it a go-to imputation method for a wide range of data types and dimensionalities.

Problem

• Imputing missing data using flow matching
• Comparing CFMI with classical and modern imputation methods
• Applying CFMI for zero-shot time-series data imputation

Innovation

• Combines continuous normalising flows and flow-matching
• Uses shared conditional modelling for missing data
• Outperforms traditional and modern imputation techniques