Deep time series models for scarce data

Time series data have grown at an explosive rate in numerous domains and stimulated a surge of time modeling research. A comprehensive comparison different models, for considered analytics task, provides useful guidance on model selection practitioners. Data scarcity is universal issue that occurs vast range problems, due to the high costs associated with collecting, generating, labeling as well some quality issues such missing data. In this paper, we focus temporal classification/regression problem attempts build mathematical mapping from multivariate inputs discrete class label or real-valued response variable. For specific problem, identify two types scarce data: small samples sparsely irregularly observed covariates. Observing all existing works are incapable utilizing sparse proper building, propose called functional multilayer perceptron (SFMLP) handling sparsity The effectiveness proposed SFMLP under each scarcity, conventional deep sequential learning models (e.g., Recurrent Neural Network, Long Short-Term Memory), investigated through arguments numerical experiments.