LAUSR

Using a deep neural network model, we show that predicting an accurate topological invariant is possible from one-dimensional Hamiltonians, whose topological invariant is the winding number. Given a set of Hamiltonians in momentum space as the input, the deep neural network can predict the topological invariant with high accuracy. We found that, upon optimization of the weight parameters, the deep neural network can accurately predict the topological invariant, if the training data size is sufficiently large. This means that extracting actionable insights from massive amounts of data is possible in the present deep neural network model.