GAIN: Missing Data Imputation using Generative Adversarial Nets

By Jinsung Yoon, James Jordon, Mihaela van der Schaar. 2018

1,410 citations recorded by the source corpus.

We propose a novel method for imputing missing data by adapting the well-known Generative Adversarial Nets (GAN) framework. Accordingly, we call our method Generative Adversarial Imputation Nets (GAIN). The generator (G) observes some components of a real data vector, imputes the missing components conditioned on what is actually observed, and outputs a completed vector. The discriminator (D) then takes a completed vector and attempts to determine which components were actually observed and which were imputed. To ensure that D forces G to learn the desired distribution, we provide D with some additional information in the form of a hint vector. The hint reveals to D partial information about the missingness of the original sample, which is used by D to focus its attention on the imputation quality of particular components. This hint ensures that G does in fact learn to generate according to the true data distribution. We tested our method on various datasets and found that GAIN significantly outperforms state-of-the-art imputation methods.

Subfields: Artificial Intelligence, Computer Vision and Pattern Recognition

How researchers see this paper

The researchers whose trusted networks rank it highest, strongest advocate first. Every researcher gets a different order.

In Artificial Intelligence

In Computer Vision and Pattern Recognition

Citations in the corpus

Learning Neural Causal Models from Unknown Interventions (2019, methodology): “Neural network methods equipped with learned masks, such as [46, 47, 48, 49], exist in the literature, but only a few [44] have been adapted to causal inference.”
GP-VAE: Deep Probabilistic Time Series Imputation (2019, background): “Another avenue of research in this area uses deep learning techniques, such as variational autoencoders (VAEs) [1, 10, 35, 37] or generative adversarial networks (GANs) [30, 47].”
Flow Models for Arbitrary Conditional Likelihoods (2019, methodology): “For MICE, MissForest and GAIN, we do not have access to the exact likelihoods.”
MIWAE: Deep Generative Modelling and Imputation of Incomplete Data Sets (2019, methodology): “Yoon et al. (2018) considered a way to do so by building on the generative adversarial networks (GANs) of Goodfellow et al. (2014).”
Handling Incomplete Heterogeneous Data using VAEs (2018, methodology): “• GAIN: Generative adversarial network for missing data imputation (Yoon et al., 2018), which uses MSE as a loss function for numerical variables, and cross-entropy for binary variables.”
Variational Autoencoder with Arbitrary Conditioning (2018, methodology): “After that we impute missing features using MICE (Buuren & Groothuis-Oudshoorn, 2010), MissForest (Stekhoven & Bühlmann, 2011), GAIN (Yoon et al., 2018) and VAEAC trained on the observed data.”
Multivariate Time Series Imputation with Generative Adversarial Networks (2018, background): “There also exists a few works that use GAN to impute the missing values [46].”
How artificial intelligence and machine learning can help healthcare systems respond to COVID-19 (2020, methodology): “They can anticipate a patient’s future healthcare needs, and can also handle irregularly collected data and impute missing features (Yoon et al. 2018a, b), as well as inferring important predictors of patient outcomes (e.g. specific comorbidities that are more likely to worsen the prognosis of a…”
Digital Twin: Values, Challenges and Enablers From a Modeling Perspective (2019, methodology): “The method has huge potential in improving the data quality like upscaling image resolution [127], denoising [128], filling missing data [129] all of which are relevant in the context of digital twins.”
SAITS: Self-Attention-based Imputation for Time Series (2022, background): “imputation models, which are not in the scope of this paper, can be referred to [22, 23, 24, 25, 26, 27, 28, 29].”