Yousaf Khaliq, a Computational and Data Science doctoral student at MTSU, presents an overview of his research and dissertation work in preparation for his final doctoral defense in the CDS program.
This works presents a modular approach to using near-term quantum computing inside hybrid quantum–classical workflows that remain anchored to classical correctness guarantees and practical performance diagnostics. Instead of fully quantum end-to-end pipelines, parameterized quantum circuits (PQCs) are treated as trainable subroutines embedded within classical learning and inference scaffolds, with the goal of identifying when quantum structure yields tangible, repeatable benefits under realistic NISQ constraints. Two case studies illustrate this framework: (1) credit-card fraud detection under extreme class imbalance, where circuit entanglement topology is used as an inductive bias aligned with empirical feature dependencies and evaluated via precision–recall behavior; and (2) probabilistic inference via the Adaptive Quantum Coordinate Field sampler, which learns a coarse categorical model of state space with a PQC and injects directed independence proposals into a classical Metropolis–Hastings mixture kernel. Experiments on warped bimodal and higher-dimensional mixture targets show improved multimodal exploration and effective sample size in metastable regimes relative to strong classical baselines. This study concludes with a quantifiable speed–fidelity trade-off from asynchronous/parallel quantum proposal execution—often improving wall-clock efficiency but introducing failure modes tied to proposal staleness.
Check out the webinar here.