Research

Trainability of quantum machine-learning models: Quantum models are known to suffer from the “barren plateau” problem, in which loss-function gradients decrease exponentially with the number of qubits. Conversely, in the overparameterized regime—where the number of trainable parameters exceeds the amount of data—generalization performance can improve abruptly through the “double-descent” phenomenon. We investigate trainability and generalization using both theoretical analysis and numerical simulations.

Learning quantum dynamics and circuit properties: We study tasks that predict expectation values of observables from Hamiltonian time evolution with unknown parameters, investigating quantum advantage in regimes where classical learning is difficult.

Ground-state simulation with near-term quantum computing: As a potential demonstration of quantum advantage, we are simulating ground-state energy measurements in lattice gauge theories. Using a 156-qubit Heron processor and Krylov-subspace sampling, combined with error mitigation on the Miyabi supercomputer, we pursue high-precision ground-state energy measurements.

Ground-state simulation with fault-tolerant quantum computing: For early FTQC, we are developing time-evolution-based ground-state preparation methods that avoid complex controlled gates and large numbers of ancillary qubits.

We also develop gauge-invariant quantum sampling methods for finite-temperature and finite-density systems, and study quantum simulations of topological defects and fermion scattering related to the cosmological matter–antimatter asymmetry.

Symmetry-based error-correction algorithms: We study quantum error correction that exploits local gauge symmetries, a defining feature of gauge theories.

Error correction with concatenated codes: To improve error-estimation accuracy while reducing implementation costs on quantum hardware, we develop theoretical designs for concatenated codes that combine autonomous error-correcting codes with stabilizer codes.

Quantum-noise estimation: Accurate noise-channel estimation is essential for error correction, but improving estimation accuracy typically increases computational cost. We investigate efficient, high-precision methods that retain non-Pauli components, such as coherent noise.

Superconducting devices: Superconducting qubits couple strongly to photons and enable nondestructive measurements with ultra-low energy thresholds of order μeV, making them powerful probes of dark matter. We are pursuing the DarQ experiment, which uses superconducting qubits to search for wave-like dark matter, including axions and dark photons.

Quantum algorithms for enhanced sensitivity: By using entangled qubit arrays and quantum-interference-based signal amplification, we aim to achieve measurement sensitivities approaching the Heisenberg limit.