Missing Data Imputation using Neural Cellular Automata

When working with tabular data, missingness is always one of the most painful problems. Throughout many years, researchers have continuously explored better and better ways to impute missing data. Recently, with the rapid development evolution in machine learning and deep learning, there is a new trend of leveraging generative models to solve the imputation task. While the imputing version of famous models such as V ariational Autoencoders or Generative Adversarial Networks were investigated, prior work has overlooked Neural Cellular Automata (NCA), a powerful computational model. In this paper, we propose a novel imputation method that is inspired by NCA. We show that, with some appropriate adaptations, an NCA-based model is able to address the missing data imputation problem. We also provide several experiments to evidence that our model outperforms state-of-the-art methods in terms of imputation error and post-imputation performance. Introduction There is no doubt that data plays a crucial role in this modern world. In numerous business and scientific applications, data is the foundation for decision-making process, enabling experts to detect noticeable patterns and take advantage of them. One of the most common types of data is tabular data, which presents in almost every domains from economics, finance to healthcare, demography. Being organized in structured rows and columns, one can straightforwardly apply statistical methods, perform calculations and draw meaningful insights from this data. Moreover, many machine learning algorithms, especially those used in supervised learning tasks, are designed to work optimally on tabular data.