A memristive computational neural network model for time-series processing

In this work, we introduce a novel computational framework inspired by the physics of memristive devices and systems, which we embed into the context of Recurrent Neural Networks (RNNs) for time-series processing. Our proposed memristive-friendly neural network architecture leverages both the principles of Reservoir Computing (RC) and fully trainable RNNs, providing a versatile platform for sequence learning. We provide a mathematical analysis of the stability of the resulting neural network dynamics, identifying the role of crucial RC-based architectural hyper-parameters. Through numerical simulations, we demonstrate the effectiveness of the proposed approach across diverse regression and classification tasks, showcasing performance that is competitive with both traditional RC and fully trainable RNN systems. Our results highlight the scalability and adaptability of memristive-inspired computational architectures, offering a promising path toward efficient neuromorphic computing for complex sequence-based applications.