Insights on Training Neural Networks for QUBO Tasks
Abstract
Current hardware limitations restrict the potential when solving quadratic unconstrained binary optimization (QUBO) problems via the quantum approximate optimization algorithm (QAOA) or quantum annealing (QA). Thus, we consider training neural networks in this context. We first discuss QUBO problems that originate from translated instances of the traveling salesman problem (TSP): Analyzing this representation via autoencoders shows that there is way more information included than necessary to solve the original TSP. Then we show that neural networks can be used to solve TSP instances from both QUBO input and autoencoders' hiddenstate representation. We finally generalize the approach and successfully train neural networks to solve arbitrary QUBO problems, sketching means to use neuromorphic hardware as a simulator or an additional co-processor for quantum computing.
Cite
@article{arxiv.2004.14036,
title = {Insights on Training Neural Networks for QUBO Tasks},
author = {Thomas Gabor and Sebastian Feld and Hila Safi and Thomy Phan and Claudia Linnhoff-Popien},
journal= {arXiv preprint arXiv:2004.14036},
year = {2020}
}
Comments
6 pages, 5 figures, accepted at the 1st International Workshop on Quantum Software Engineering (Q-SE 2020) at ICSE 2020 and to be published in the corresponding proceedings