Quantum algorithm for twin extreme learning machine

Abstract

Twin extreme learning machine (TELM) is a powerful learning algorithm, which aims at learning two nonparallel hyperplanes for data classification. However, classical TELM algorithm becomes computationally expensive when it involves big data sets. In this paper, we devise a quantum TELM algorithm to address this issue. Specifically, we first utilize the quantum amplitude estimation algorithm to prepare the desired input states and then call the quantum linear systems of equations, which adopts block-encoding technique, to obtain the model parameters in the training process. Then we invoke the swap test to estimate the distances from a new data point to the two hyperplanes and then make a classification in the prediction stage. The final complexity analysis demonstrates that our algorithm has an exponential speedup under certain conditions over the classical counterpart.