Quantum computer that learns from its own errors

Google Research, in collaboration with Google DeepMind, published a paper in Nature titled "Reinforcement learning control of quantum error correction" demonstrating a reinforcement learning (RL) framework that enables a quantum computer to continuously adapt to drift during long computations without stopping. The RL agent learns from binary error detection events generated by quantum error correction (QEC) to dynamically steer thousands of control parameters such as frequencies, amplitudes, and phases. The approach was validated on the Willow superconducting processor, where RL steering improved logical stability of the error-correcting code 3.5-fold. Fine-tuning with RL after exhaustive expert calibration further reduced the logical error rate by 20%. The combined technologies achieved record low logical error rates: fewer than one per thousand error correction cycles in the surface code, and one per hundred in the color code. Numerical simulations with hundreds of qubits showed that the number of required RL training iterations is independent of system size.