AI and Quantum Computing

Quantum computing offers enormous potential for improving and accelerating the solution of problems in the areas of simulation, optimization and machine learning. In particular, variational quantum circuits - a special class of quantum algorithms - have shown advantages in the field of machine learning in certain data regimes. When used as a machine learning model, variational quantum circuits can learn a problem just as well, but often with less data, than classical approaches such as neural networks.

At 5G Connect Advanced, it will be shown how variational quantum circuits are combined with a special form of machine learning: reinforcement learning. Here, an algorithm (“the agent”) attempts to solve a specific problem (“the environment”) and find the best possible solution strategy.

The algorithm “Meta-Learning Restless Multi-armed Bandits for sample efficient online inference” will be presented at 5G Connect Advanced.

This algorithm uses a special coordinate system based on codebook entries instead of a normal 3D coordinate system. This allows it to directly determine the correct beam direction without having to localize the receiver in X-Y-Z coordinates. This saves expensive localization steps and makes the algorithm more robust against physical obstacles such as cars or trees.

Machine learning enables localization in complex, angled environments where conventional localization methods reach their limits. Neural networks use environmental information that is encoded in the radio signals and that conventional methods cannot process. The uncertainty estimation of neural networks and unsupervised learning can therefore enable reliable and efficient area-wide localization.