Projects

Speculation strategies that detect and mitigate leakage errors in QEC codes — 16% better logical error rate, 2–3× less leakage accumulation.

Hardware-efficient ML architectures for superconducting qubit readout — 16% accuracy gain, matched filters under 8% FPGA resources, 6.6% multi-level improvement.

Modular, scalable ML models for neutral-atom qubit readout — 70× smaller models than state of the art, plus image denoising for fast, accurate readout.

Crosstalk-based side-channel attacks on multi-tenant NISQ hardware — deciphering a victim’s circuit with graph neural networks.

A unified adaptive ML model (GAIL imitation learning) for Rz/unitary synthesis and quantum compilation across dynamic hardware configurations. Ongoing.