<ul><li>Recurrent Neural Networks (RNNs) are capable of learning functions on sequence data.</li><li>Reservoir computers, a class of RNNs, trained on dynamical system observations can be interpreted as embeddings.</li><li>An upper bound for the fractal dimension of the reservoir state space during training and prediction phase is established.</li><li>The fractal dimension of the subset is bounded above by the dimension of the input sequences in a nonautonomous dynamical system.</li></ul>

On the dimension of pullback attractors in recurrent neural networks

Discover more