Tensor Train Recurrent Network Language Model Prediction

Recurrent neural networks (RNN) such as long‐short‐term memory (LSTM) networks are essential in a multitude of daily tasks such as speech, language, video and multimodal learning. The shift from cloud to edge computation intensifies the need to contain the growth in size of RNNs. Current research on RNN shows that despite the performance obtained on convolutional neural networks (CNN), keeping a good performance in compressed RNNs is still a challenge. This paper shows that by incorporating informative matrix‐normal priors on the tensor weights, tensor‐compressed LSTM networks can achieve comparable performance to LSTM networks. Most literature on compression focuses on CNNs using matrix product (MPO) operator tensor trains. However, matrix product state (MPS) tensor trains have more attractive features in terms of storage reduction and computing time for prediction. The present work shows that MPS tensor trains should be at the forefront of LSTM network compression through a theoretical analysis and practical experiments on natural language processing (NLP) tasks.