Empirical Learning for Dynamical Decoupling On Quantum CPUs
Empirically Optimised Dynamical Decoupling Improves Quantum Computation.
Despite its potential, quantum computing is hindered by the fragility of quantum states, which are susceptible to ambient noise mistakes. Numerous error suppression and mitigation solutions are being studied to eliminate these defects and increase quantum calculation reliability. Dynamical Decoupling reduces quantum computation defects most efficiently and cheaply.
Suppressing Errors on Noisy Hardware
Despite advances in DD design, pulse sequences that successfully decouple computational qubits on noisy quantum devices remain a challenge. Quantum processor noise is complicated and dynamic, therefore theoretically-derived DD sequences often fail in practice. This requires a more flexible and hardware-sensitive DD implementation.
Empirical Optimisation: A New DD Method
Recent discoveries have introduced a powerful paradigm: empirical optimisation of DD sequences. Learning algorithms or combinatorial optimisation are used to empirically find device-tailored DD sequences. Instead than using theoretical models, this method discovers optimal approaches using quantum hardware experiments.
Empirical learning is used to optimise DD (GADD), inspired by genetic algorithms. In particular, this strategy has optimised DD procedures for IBM's superconducting qubit-based quantum processors. GADD simulates natural selection by enhancing Dynamical Decoupling (DD) sequences based on how well they suppress quantum device errors.
Outstanding Performance and Generalisability
Empirical optimisation yielded excellent results. We found that empirically taught DD approaches suppress errors better than canonical sequences in all experimental settings. These experimentally optimised sequences reduce superconducting qubit noise better than theoretically derived DD sequences and traditional decoupling sequences like CPMG, XY4, and UR6.
The harder the computer problem, the better these experimentally learnt solutions work. Error suppression improves with problem size and circuit complexity. This discovery is significant as quantum algorithms get more complex and require more qubits and deeper circuits.
Advantages of this empirical learning approach include stability and generalisability.
The approaches identified provide long-term performance without retraining. Maintaining peak performance reduces operational expenditures.
The optimisation approach is scalable and successful for increasingly complex quantum systems since these empirically taught methods can generalise to bigger circuits when trained on modest sub-circuit architectures. The methodology also finds time-constant ways as circuit width and depth increase.
Practical Uses and Benchmarking
Empirically optimised Dynamical Decoupling (DD) performs well on many quantum techniques and scales:
Mirror randomised benchmarking on 100 qubits was examined.
This prepared 50 qubits for GHZ states.
It improved the 27-qubit Bernstein-Vazirani algorithm.
This method can increase near-term quantum gadget performance, as shown by these investigations. IBM has led this study and may have combined these results with Qiskit. IBM prioritises quantum computing and software.
Bigger context and future implications
Dynamic decoupling is a simple error suppression method. Combining it with empirical learning makes quantum computation more resilient. This study aligns with quantum error mitigation efforts including adaptive Dynamical Decoupling (DD) frameworks and context-aware compilation to reduce crosstalk and correlated noise. It emphasises the importance of a comprehensive strategy that uses hardware and algorithm design to enable reliable and high-quality near-term quantum algorithm execution.
