Quantum Recurrent Neural Networks for Multivariate Time Series Prediction

Some Machine Learning algorithms, like Recurrent Neural Networks (RNNs), analyse time series to predict unknown values of variables in a complex system. When dealing with multi- layer networks and broad series, some issues, such as overfitting or memory losses, arise. Several approaches intend to address them, for example, the Long Short-Term Memory (LSTM) cell. Despite these approaches, learning from multivariate-complex systems is still a challenge and requires networks with many non-linear terms, expensive to compute on classical devices. Quantum Computation emerges as a promising approach to tackle complex problems more efficiently since it allows to compute non-linear terms in a high dimensional space without spending exponential resources. We propose a Quantum RNN (QRNN) model as a first step towards multivariate time series forecasting. The core of QRNN is a parameterized quantum circuit that iteratively exchanges information, but, at the same time, it keeps memory from past data

keywords: