Insights on Training Neural Networks for QUBO Tasks
This addresses computational bottlenecks for researchers and practitioners in optimization and quantum computing, though it appears incremental as it adapts existing neural network techniques to a new context.
The paper tackles the hardware limitations in solving quadratic unconstrained binary optimization (QUBO) problems via quantum methods by training neural networks, showing they can solve TSP instances from QUBO input and autoencoder representations, and generalize to arbitrary QUBO problems with potential neuromorphic hardware applications.
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.