Towards quantum amenable Bayesian networks: classical transformation to facilitate quantum inference

Inference in Bayesian networks has been an interesting area of research due to the ability of Bayesian networks in accurately modeling uncertainty in complex use-cases. Performing inference by utilizing quantum computing is possible due to recent works in the area and readily available quantum hardware. However, there are certain limits in terms of qubits when it comes to large-scale Bayesian network inference. We propose a novel segmentation-based approach to perform piece-meal inference on Bayesian networks, where individual segments are run as an independent quantum circuit and act as a prior circuit for downstream segments. In the case of larger and more complex individual segments, we propose Quantum Amenable Classical Transformation (QACT) to channel the likelihoods through a newly added intermediate node, thereby reducing the complexity of the network. Experimental results on various benchmark Bayesian networks demonstrate that the combined use of segmentation and QACT significantly reduces the qubit requirements compared to the existing QBN approach without any statistically significant impact on the inference accuracy.