Thermalizing stochastic programs
Compiling general stochastic programs into thermodynamic kernels that can be sampled on probabilistic hardware.
I study computation in which physical stochastic dynamics are part of the algorithm. The goal is to co-design probabilistic hardware, stochastic programs, and compilation methods so that useful distributions can be sampled with low energy and control overhead.
I work on calibration and control and on the interface between algorithms and thermodynamic hardware. Parametrized Stochastic Circuits provide a compositional representation of typed stochastic kernels. Torx implements this representation in JAX. Thermalizers compiles these kernels to hardware-native energy-based models and provides methods to control accumulated compilation error.
Compiling general stochastic programs into thermodynamic kernels that can be sampled on probabilistic hardware.
A gate-based intermediate representation and JAX framework for constructing, executing, and differentiating stochastic programs.
My previous work developed and implemented quantum algorithms tailored to compute quantities of interest within the coherence budget of noisy hardware. Topics range from simulations of quantum many-body systems, to high-energy physics phenomena, quantum thermodynamics and quantum optics. Highlights include Krylov-subspace diagonalisation of spin lattices, the first experimental measurement of conformal-field-theory central charge on a universal processor, and early demonstrations that qubit-level matched filtering can recover astrophysical signals at classical signal-to-noise ratios.
Replacing controlled time evolution with uncontrolled evolution to reduce the two-qubit-gate cost of phase estimation.
Fast simulation of fermionic quantum circuits using particle-number and spin symmetries.
Diagonalizing large many-body Hamiltonians via shallow Trotter circuits and classical post-processing.
Diagonalizing large many-body Hamiltonians via shallow Trotter circuits and sampling in the computational basis.
Enhancing classification of healthcare data with projected quantum kernels.
Measuring the central charge of a quantum system on a quantum processor.
Reliable algorithmic results on noisy quantum processors require hardware-aware error suppression and mitigation. I developed tools for autonomous calibration of control pulses using deep reinforcement learning and black-box optimization, compiler techniques that reduce crosstalk using dynamical decoupling, and qubit-selection protocols based on sparse-noise tomography. I also studied hybrid quantum-classical methods, including tensor-network-assisted multiproduct formulas and quantum-enhanced Pauli propagation. Across these projects, the aim was to suppress or correct dominant errors while keeping sampling overhead manageable.
Using noisy quantum data to improve approximate classical Pauli-path simulation without explicit noise characterization.
Compressing time-evolution circuits using multiproduct formulas assisted by tensor networks.
Toolkit for modeling device noise and reconstructing unbiased expectation values using quasi-probability techniques.
Choosing the best qubits using sparse noise tomography or aggregated error metrics.
Mitigating correlated noise at compile time via targeted dynamical decoupling.
Using deep reinforcement learning and black-box optimization to autonomously calibrate high-fidelity gates.
At the heart of many superconducting devices is a tunable qubit-cavity coupling. My doctoral work focused on non-adiabatic effects, the dynamical Lamb and Casimir effects, arising during fast qubit driving. Modeling of superconducting circuits shows that these effects can be used to generate entanglement and perform ultra-fast quantum gates.
Entanglement generation in non-stationary cavity QED via the dynamical Lamb effect.
For citation metrics and newly added preprints, see my Google Scholar profile.