Graph neural network initialisation of quantum approximate optimisation

• DOI: 10.22331/q-2022-11-17-861
• Publication Date: 2022-11-17T15:09:55Z
• Authors: Nishant Jain; Brian Coyle; Elham Kashefi; Niraj Kumar
• Keywords: Benchmark (surveying); Feature (linguistics)

Approximate combinatorial optimisation has emerged as one of the most promising application areas for quantum computers, particularly those in near term. In this work, we focus on approximate algorithm (QAOA) solving MaxCut problem. Specifically, address two problems QAOA, how to initialise algorithm, and subsequently train parameters find an optimal solution. For former, propose graph neural networks (GNNs) a warm-starting technique QAOA. We demonstrate that merging GNNs with QAOA can outperform both approaches individually. Furthermore, enables warm-start generalisation across not only instances, but also increasing sizes, feature straightforwardly available other methods. training test several optimisers problem up 16 qubits benchmark against vanilla gradient descent. These include aware/agnostic machine learning based/neural optimisers. Examples latter reinforcement meta-learning. With incorporation these initialisation toolkits, be solved using end-to-end differentiable pipeline.