KIP-Veröffentlichungen

Jahr 2020
Autor(en) Manuel Rudolph
Titel Exploring and Benchmarking Quantum-assisted Neural Networks with Qubit Layers
KIP-Nummer HD-KIP 20-84
KIP-Gruppe(n) F17,F21
Dokumentart Masterarbeit
Keywords (angezeigt) Quantum Machine Learning, Cold Atoms, Generative Models
Abstract (de)

The aim of this work is to explore practical implementations of Quantum and Quantum-assisted Machine Learning algorithms and benchmark potential bene- ts of utilizing quantum phenomena in Quantum-assisted Neural Networks with qubit layers. Two known approaches of generative Quantum Machine Learning algorithms are revised to demonstrate the encoding capability and sampling bene ts of qubits. As one possible extension of those generative models, the Quantum-assisted Generator (QaG) is presented which implements a qubit layer coupled to subsequent classical layers. The QaG provides promising indications that quantum phenomena may enhance performance of a generative neural network. This work also considers Hamiltonian-based and gate-based implementations of quantum-assisted Autoencoders. Using an e ective hybrid training approach consisting of simultaneous application of conventional backpropagation and blackbox optimization, we show that the Hamiltonian-based Autoencoder is able to learn the MNIST data set with good generalization properties. Several approaches to extend the presented quantum-assisted algorithms are proposed to further investigate potential bene ts of utilizing quantum systems in combination with Arti cial Neural Networks.

Abstract (en)

The aim of this work is to explore practical implementations of Quantum and Quantum-assisted Machine Learning algorithms and benchmark potential bene- ts of utilizing quantum phenomena in Quantum-assisted Neural Networks with qubit layers. Two known approaches of generative Quantum Machine Learning algorithms are revised to demonstrate the encoding capability and sampling bene ts of qubits. As one possible extension of those generative models, the Quantum-assisted Generator (QaG) is presented which implements a qubit layer coupled to subsequent classical layers. The QaG provides promising indications that quantum phenomena may enhance performance of a generative neural network. This work also considers Hamiltonian-based and gate-based implementations of quantum-assisted Autoencoders. Using an e ective hybrid training approach consisting of simultaneous application of conventional backpropagation and blackbox optimization, we show that the Hamiltonian-based Autoencoder is able to learn the MNIST data set with good generalization properties. Several approaches to extend the presented quantum-assisted algorithms are proposed to further investigate potential bene ts of utilizing quantum systems in combination with Arti cial Neural Networks.

bibtex
@mastersthesis{MasterRudolph,
  author   = {Manuel Rudolph},
  title    = {Exploring and Benchmarking Quantum-assisted Neural Networks with Qubit Layers},
  school   = {Universität Heidelberg},
  year     = {2020},
  type     = {Masterarbeit}
}
Datei pdf
KIP - Bibliothek
Im Neuenheimer Feld 227
Raum 3.402
69120 Heidelberg