SAMOP 2023 – scientific programme

Parts | Days | Selection | Search | Updates | Downloads | Help

QI: Fachverband Quanteninformation

QI 3: Quantum Machine Learning

QI 3.2: Talk

Monday, March 6, 2023, 11:30–11:45, B305

The application of quantum neural networks in function approximation — •David Kreplin and Marco Roth — Fraunhofer IPA, Nobelstraße 12, 70569 Stuttgart, Deutschland

Approximating functions by parameterized quantum circuits is a promising application for quantum computing, since the repetitive encoding of the input data can result in an exponentially growing complexity of the function. In the literature, this approach is often described as Quantum Neural Networks (QNNs), since it can be similarly utilized as classical artificial neural networks.

In this talk, we show how an efficient and general function approximation can be realized by a QNN. We discuss the construction, training, and the application of the QNN with the example of solving a differential equation based model of a hydrogen electrolyzer and benchmark the results against classical neural networks.

A particular focus in this talk will be on the unavoidable noise that results from the finite sampling of the quantum state. This so-called shot noise strongly degrades the training process and yields a noisy outcome of the QNN. We discuss how that shot noise can be strongly reduced during the training of the QNN by an additional regularization term. This not only reduces the noise in the final function but also simplifies the training process on shot based simulators or real devices. Finally, we present results from the real quantum computing hardware and we reflect on the obstacles that we currently face in training such QNNs on the real backends.