Sparsified Kolmogorov-Arnold Networks for Interpretable Quantum State Tomography

The article presents a sparsified Kolmogorov-Arnold Network (KAN) applied to quantum state tomography, demonstrating its capability as both a regressor and an interpretable reconstruction rule that aligns with known Pauli structures. The study utilizes a three-qubit GHZ-family benchmark to reconstruct key variables, achieving exact recovery of the top-12 GHZ-relevant Pauli measurements under various noise conditions. This approach enhances pathway-level interpretability within neural models, allowing practitioners to audit learned reconstruction rules against established physical frameworks, which is critical for advancing reliable quantum machine learning applications.