Regression on sequential data is an important and extensively studied topic with a broad range of applications in the machine learning literature. However, in most real-life applications, these data sequences usually suffer from missing data problems due to various reasons such as sensor failures and communication delays, which severally degrade the performance of most, if not all, machine learning algorithms. The proposed architecture employs different LSTM networks tuned to the specific presence-patterns and, therefore, inherently captures the underlying dynamics without using ad hoc imputation methods or statistical assumptions on the data. As imputation-based methods may experience performance decreases due to the mismatch between the imputation model and the data, our algorithm does not suffer from this problem since we do not make any statistical or structural assumptions on the missing data. In our experiment results, owing to not having any such assumptions, our approach provid...

We study regression (or prediction) of sequential data, which may have missing entries and/or different lengths. This problem is heavily investigated in the machine learning literature since such missingness is a common occurrence in most real-life applications due to data corruption, measurement errors, and similar. To this end, we introduce a novel hierarchical architecture involving a set of long short-term memory (LSTM) networks, which use only the existing inputs in the sequence without any imputations or statistical assumptions on the missing data. To incorporate the missingness information, we partition the input space into different regions in a hierarchical manner based on the “presence-pattern” of the previous inputs and then assign different LSTM networks to these regions. In this sense, we use the LSTM networks as our experts for these regions and adaptively combine their outputs to generate our final output. Our method is generic so that the set of partitioned regions (pre...