Automated and Predictive Performance Evaluation in compleX environments for Fault Tolerant Quantum Computing (APPEX FTQC)
Abstract
The APPEX consortium will develop new approaches for quantum characterization, verification and validation (QCVV), which bridge the gap between fault-tolerant analyses and laboratory-realistic noise environments, and leverage the most modern knowledge in deep learning and quantum control to enable efficiency and autonomy at large scales. In realistic laboratory environments, typical noise and error processes frequently to violate the strict assumptions that underpin the application of QCVV outputs to fault-tolerance analyses. This problem is amplified as systems grow and the space of potential correlations grows exponentially Ð as does the space of potential measurables in QCVV. In combination these effects erode trust in QCVV outputs when extrapolating to large-scale systems. We will address this and related challenges by augmenting QCVV protocols with a new class of proxy measures for performance-limiting microscopic error details. Through a combination of efficient model reduction and powerful deep learning techniques we will restore trust in QCVV outputs by focusing on efficiently capturing the microscopic details which can dramatically alter the interpretation of QCVV. These will address noise correlations, cross-talk, drifts, gate-contextuality, and nonlinear-control-transduction faced in real devices. By appropriately accounting for the relevant microscopic details of the noise processes underlying hardware error rates, we will ensure that QCVV routines directly output information that is predictive for fault-tolerance without being overly pessimistic or unreasonably unphysical. Core to our approach is the integration of novel machine learning routines that augment the efficiency of hardware characterization Ð even in the presence of uncertainty in the Hamiltonian representation of the system and imperfect measurements. Bringing insights from both adaptive learning/robotic control and deep reinforcement learning we will enable the extension of these novel QCVV routines to large systems by reducing the number of required measurements by at least an order of magnitude in mesoscale QCVV and achieving polynomial scaling of measurement counts with system size. These procedures work by exploiting spatio-temporal correlations and ensuring the data inference technique is aware of the relationships between measurements on proximal qubits. We will also perform the first rigorous study of the interface of parameterized QCVV approaches (even with greater flexibility than today s routines) with deep reinforcement learning (RL) approaches that are proven to be effective in high-dimensional spaces. We believe that integration of RL with QCVV is an essential step towards large-scale fault-tolerant devices. Our effort is centered around a tight integration of theoretical analysis and scalable algorithmic design with two complementary experimental efforts in trapped-ion and superconducting qubits Ð a combination which allows for efficient co-design and direct cross-platform testing of all newly developed methods at intermediate scales of 10-20 qubits. Using these cutting edge platforms we will experimentally demonstrate the predictive power of our newly developed QCVV suite in algorithmic implementations and extrapolate the necessary noise/error improvements needed to achieve fault tolerance in large scale systems of 50 or more physical qubits. Our world-leading consortium is poised to deliver major new insights to the field of QCVV in realistic large-scale systems.
Document Details
- Document Type
- DoD Grant Award
- Publication Date
- Oct 22, 2020
- Source ID
- W911NF2110003
Entities
People
- Michael J. Biercuk
Organizations
- Army Contracting Command
- National Security Agency
- University of Sydney