Iterative Imputation of Missing Data Using Auto-Encoder Dynamics

摘要: This paper introduces an approach to missing data imputation based on deep auto-encoder models, adequate high-dimensional exhibiting complex dependencies, such as images. The method exploits the properties of vector field associated auto-encoder, which allows approximate gradient log-density from its reconstruction error, we propose a projected ascent algorithm obtain conditionally most probable estimate values. Our does not require any specialized training procedure and can be used together with model trained complete in classical way. Experiments performed benchmark datasets show that imputations produced by our are sharp realistic.