Item

Abstract

Artificial neural networks are revolutionizing science. While the most prevalent technique involves supervised training on queries with a known correct answer, more advanced challenges often require discovering answers autonomously. In reinforcement learning, control strategies are improved according to a reward function. The power of this approach has been highlighted by spectactular recent successes, such as playing Go. So far, it has remained an open question whether neural-network-based reinforcement learning can be successfully applied in physics. Here, we show how to use this method for finding quantum feedback schemes, where a network-based "agent" interacts with and occasionally decides to measure a quantum system. We illustrate the utility by finding gate sequences that preserve the quantum information stored in a small collection of qubits against noise. This specific application will help to find hardware-adapted feedback schemes for small quantum modules while demonstrating more generally the promise of neural-network based reinforcement learning in physics.