QIBONN: A Quantum-Inspired Bilevel Optimizer for Neural Networks on Tabular Classification

QIBONN, a newly introduced Quantum-Inspired Bilevel Optimizer for Neural Networks, addresses the challenges of hyperparameter optimization (HPO) in neural networks applied to tabular data. The framework utilizes a unique qubit-based representation to unify feature selection, architectural hyperparameters, and regularization. By integrating deterministic quantum-inspired rotations with stochastic qubit mutations, QIBONN optimizes the balance between exploration and exploitation within a fixed evaluation budget. Systematic experiments conducted using an IBM-Q backend under single-qubit bit-flip noise have shown that QIBONN is competitive with established methods, including classical tree-based approaches and other HPO algorithms, across 13 real-world datasets. This development is significant as it enhances the efficiency and effectiveness of machine learning applications, particularly in environments where exhaustive tuning is impractical.