Jamie Heredge, Maxwell West, Lloyd Hollenberg, Martin Sevior (May 28 2024).
Abstract: We introduce several novel probabilistic quantum algorithms that overcome the normal unitary restrictions in quantum machine learning by leveraging the Linear Combination of Unitaries (LCU) method. Among our contributions are quantum native implementations of Residual Networks (ResNet); demonstrating a path to avoiding barren plateaus while maintaining the complexity of models that are hard to simulate classically. Furthermore, by generalising to allow control of the strength of residual connections, we show that the lower bound of the LCU success probability can be set to any arbitrary desired value. We also implement a quantum analogue of average pooling layers from convolutional networks. Our empirical analysis demonstrates that the LCU success probability remains stable for the MNIST database, unlocking a potential quadratic advantage in terms of image size compared to classical techniques. Finally, we propose a general framework for irreducible subspace projections for quantum encoded data. Using this, we demonstrate a novel rotationally invariant encoding for point cloud data via Schur-Weyl duality. We also show how this framework can be used to parameterise and control the amount of symmetry in an encoding; demonstrating improved classification performance for partially permutation invariant encoded point cloud data when compared to non-invariant or fully permutation invariant encodings. These new general algorithmic frameworks are all constructed under the same LCU method, suggesting that even more novel algorithms could be achieved by utilising the LCU technique.