Our paper, “Phase-estimation readout for variational quantum classifiers: spectrum and benchmarks”, co-authored with Petro Chynnyk, a PhD student at the Institute for Applied Systems Analysis of Igor Sikorsky Kyiv Polytechnic Institute, has been accepted for publication in Academia Quantum (Academia.edu Journals).
In this work, we propose and study a variational quantum classifier whose output is read from a quantum phase-estimation register, showing that its response as a function of the input data is a trigonometric polynomial whose maximum harmonic order grows exponentially with the size of the estimation register. We prove when this bound is attained and when it is not, and explain where the model’s nonlinearity comes from.
We tested our theoretical findings in a large computational experiment comprising 7,680 training runs across several classification tasks, with rigorous statistical analysis of the results. On a three-class task, our model with a universal single-qubit block performed best among twelve quantum circuits on each of twenty data splits.
We also examine the limits of our approach, acknowledging that a classical support vector machine remains more accurate and that our architecture offers no advantage in parameter count. Its advantage lies in the structure of its frequency spectrum rather than in resource savings, a distinction we make explicit in our analysis.