Predicting Sequential Data with LSTMs Augmented with Strictly 2-Piecewise Input Vectors

Recurrent neural networks such as Long-Short Term Memory (LSTM) are often used to learn from various kinds of time-series data, especially those that involved long-distance dependencies. We introduce a vector representation for the Strictly 2-Piecewise (SP-2) formal languages, which encode certain kinds of long-distance dependencies using subsequences. These vectors are added to the LSTM architecture as an additional input. Through experiments with the problems in the SPiCe dataset (Balle Pigem et al., 2016), we demonstrate that for certain problems, these vectors slightly—but significantly—improve the top-5 score (normalized discounted cumulative gain) as well as the accuracy as compared to the LSTM architecture without the SP-2 input vector. These results are also compared to an LSTM architecture with an input vector based on bigrams.