Bayesian network (BN) inference—computing posterior probabilities given evidence—is a core task in probabilistic reasoning, but it becomes computationally expensive as networks grow in size or treewidth increases. Quantum-accelerated BN inference explores whether quantum algorithms and quantum circuit representations can provide practical advantages for approximate inference and sampling, while still making realistic assumptions about data access, noise, and limited quantum resources. This project will investigate how a classical BN can be transformed into a quantum-amenable representation (e.g., a quantum circuit that supports sampling from the joint distribution or from evidence-conditioned distributions). The “realistic assumptions” focus includes explicit accounting of: (i) how conditional probability tables (CPTs) are encoded, (ii) qubit and circuit-depth requirements, (iii) the impact of hardware constraints such as connectivity and noise, and (iv) the true end-to-end runtime when classical pre-/post-processing is included. A promising direction is hybrid inference: use classical methods (variable elimination, loopy belief propagation, importance sampling, likelihood weighting) to reduce or structure the problem, and then apply quantum subroutines for sampling/estimation (e.g., amplitude amplification/estimation-inspired routines, quantum rejection sampling variants, or variational/hybrid quantum circuits) to improve sample efficiency or accuracy in targeted regimes. The deliverable should include a reproducible experimental pipeline comparing classical baselines vs quantum/hybrid approaches on benchmark Bayesian networks (small-to-medium scale), with resource estimates and clear break-even analysis.