Using Autoencoders and Automatic Differentiation to Reconstruct Missing Variables in a Set of Time Series

Abstract Existing black box modeling approaches in machine learning suffer from a fixed input and output feature combination. In this paper, a new approach to reconstruct missing variables in a set of time series is presented. An autoencoder is trained as usual with every feature on both sides and the neural network parameters are fixed after this training. Then, the searched variables are defined as missing variables at the autoencoder input and optimized via automatic differentiation. This optimization is performed with respect to the available features loss calculation. With this method, different input and output feature combinations of the trained model can be realized by defining the searched variables as missing variables and reconstructing them. The combination can be changed without training the autoencoder again. Furthermore, fully missing feature data in a set of time series can be reconstructed from the available data. Possible applications are for example large production systems, inaccessible feature data in a running application, reduced datasets due to limited data processing capabilities or perhaps the reconstruction of dead pixels in a video. The approach is evaluated on the base of a strongly nonlinear electrical component and on a door frame production system. It is working well for single features missing and generally even for multiple missing features. But a coupling between the features inside a dataset is required for reconstruction of a feature.